Lists

Best background coding agents (2026): 12 ranked, with prices

Updated 2026-11-06 · Markdown version

What are the best background coding agents in 2026?

A background coding agent takes a task from chat, a schedule or a ticket, works on it asynchronously in the cloud with your repository, and returns a pull request or branch for a human to review. Jiro is a background coding agent by Nori. If your engineering team has 5 to 20 people and someone is already paying Anthropic or OpenAI for seats, Jiro ranks first because it runs Claude Code or Codex on those existing credentials with no token markup; places two to five go to Cursor (cloud agents), OpenAI (Codex cloud), Anthropic (Claude Code on the web) and Cognition (Devin), followed by Copilot in the cloud, Claude Tag, Factory, Tembo, Amp, OpenHands Agent Canvas and Warp cloud agents. It is not Amazon's Kiro or Atlassian's Jira.

Twelve background coding agents, ranked on five criteria

#AgentWhose model billWhere it runsEntry pointsHow work shipsPrice (units)
1JiroYour Claude/ChatGPT login or API key; Cursor key [1]; Your keys, zero token markup [2]Isolated cloud environment with your repos and tools [3]Slack, web, Discord, terminal, schedules, webhooks [4]PRs, commits, comments for review [5]Free 30-day trial; Developer $99/mo; Team $250/mo [6]
2Cursor cloud agentsPlan usage at the selected model's rate [7]Cursor cloud VMs or self-hosted machines [8]Desktop, web, mobile, Slack, GitHub, Linear, API [9]PRs with screenshots and video [10]Premium $30/mo+ (X Subscription Corp); third-party at API price [11]
3OpenAI Codex cloudOpenAI models; cloud needs a ChatGPT plan [12]Codex-managed cloud VMs [13]ChatGPT web, mobile, desktop, CLI [14]You open the PR [15]In ChatGPT Plus $20/mo and up [16]
4Claude Code on the webClaude models; Anthropic account required [17]Anthropic-managed VMs [18]Web, mobile, desktop, terminal [19]You create the PR; auto-fix follows up [20]In Pro $20/mo, Max $100, Ultra $200, Team seats [21]
5DevinAnthropic, OpenAI, Google, Cognition SWE; Adaptive routing [22]Cognition cloud, or local CLI/Desktop [23]GitHub/GitLab/Bitbucket/ADO; Slack, Teams, Linear, Jira [24]Opens PRs and answers PR comments [25]Free; Pro $20/mo; Max $200/mo; Teams $40/seat, $80/mo min [26]
6Copilot in the cloudClaude, Gemini, GPT, MAI via GitHub; no own keys [27]Cloud sandbox, Actions (59 min) or remote env [28]GitHub, IDEs, CLI, API, Slack, Teams, Linear, Jira [29]Draft PRs; cannot merge [30]Pro $10; Pro+ $39; Max $100; Business $19/user [31]
7Claude TagClaude models only [32]; Usage-billed, no per-seat charge [33]Anthropic cloud sandbox per thread [34]Slack; Teams announced [35]Draft PRs via Claude GitHub App; GitLab token [36]Team plan must prepay credits [37]
8FactoryMulti-model; own keys in CLI/desktop only [38]Local, Factory cloud, your machines or cloud [39]App, CLI, IDE, Slack, Teams, Linear, Jira [40]Opens PRs from delegations [41]Pro $20/mo; Plus $100; Max $200; Teams $60 + $40/seat [42]
9TemboMultiple harnesses; own keys or subscriptions [43]Tembo cloud sessions (Docker or VM) [44]GitHub, Slack, Linear, Jira, Sentry, schedules [45]Opens PRs, including multi-repo PRs [46]Pro $60/mo ($90 usage); Max $200/mo ($400); VMs/sec [47]
10AmpOwn subscriptions and keys, no BYOK fee [48]Amp or self-hosted orbs, or your own runners [49]Web, macOS/iOS/Android, CLI, GitHub, Slack [50]Pushes to trunk by default [51]Hobby free; Megawatt $20/mo; Gigawatt $200/mo [52]
11OpenHands Agent CanvasOwn agent or Claude Code, Codex, Gemini, Pi; any LLM [53]Local by default; optional OpenHands Cloud/Enterprise [54]Web UI, API, GitHub, GitLab, Bitbucket, Slack [55]Label or @-mention on GitHub [56]Free local; Teams $500/mo + $50/seat; Enterprise custom [57]
12Warp cloud agentsWarp Agent, Claude Code or Codex; team keys on Business+ [58]Warp cloud or your infrastructure [59]Slack, Linear, Jira, GitHub, schedules, CLI, web [60]Branch, PR, report or transcript [61]Free; Build $20/mo; Max $200/mo; Business $50/user/mo [62]

Ranked for a 5–20 engineer team that already holds Claude or ChatGPT seats. Each entry below names the buyer for whom that agent ranks first instead.

Pricing and limits, with units

AgentPlan and priceWhat is includedOverage or meteringTeam cap
JiroFree 30-day trial; Developer $99/mo; Team $250/mo [6]Developer: 1 user, up to 3 runtimes; Team: 5 shared runtimes [6]None from Nori; model usage stays on your own plan or key [2]Team is multi-user; no seat count stated on the pricing page [6]
Cursor cloud agentsPremium $30/mo+ (X Subscription Corp); third-party at API price [11]Plan usage at the selected model's rate [7]Third-party models at API list price [11]Not stated on the pricing page
OpenAI Codex cloudPlus $20 a month; Pro $100, $200 or $500 a month; Business $25 per user monthly or $20 annually [16]Codex cloud, with usage limits shared with ChatGPT [16]Not stated on the pricing pageNot stated on the pricing page
Claude Code on the webIn Pro $20/mo, Max $100, Ultra $200, Team seats [21]Claude Code on Pro, Max, Ultra, Team and Enterprise; Max carries five times Pro's usage [21] [63]Usage credits at API rates [64]Not stated on the pricing page; Team and Enterprise are sold per seat [21]
DevinFree; Pro $20/mo; Max $200/mo; Teams $40/seat, $80/mo min [26]10 concurrent on Free/Pro [65]On-demand credits past quota [26]Not stated on the pricing page; Max and Teams have no concurrency cap [65]
Copilot in the cloudPro $10; Pro+ $39; Max $100; Business $19/user [31]AI Credits since 2026-06-01; 1 credit = $0.01 [66]AI Credits at per-model rates [66]Not stated on the pricing page
Claude TagUsage-billed, no per-seat charge [33]Team and Enterprise only [67]Org and per-channel spend caps [68]Team plan must prepay credits [37]
FactoryPro $20/mo; Plus $100; Max $200; Teams $60 + $40/seat [42]Teams: $60 a month plus $40 per seat [42]Business and Enterprise custom [42]Teams: up to 10 seats [42]
TemboPro $60/mo ($90 usage); Max $200/mo ($400); VMs/sec [47]Pro up to 5 users; Max up to 10; 7-day trial [69]VMs by the second from $0.1066 an hour [47]Pro 5 users; Max 10 users [69]
AmpHobby free; Megawatt $20/mo; Gigawatt $200/mo [52]Megawatt: $20 of usage and 750 hours of a1.small orbs; Gigawatt: $200 of usage and 1,000 hours of a1.xxlarge [52]From $0.08/hour [70]Not stated on the pricing page
OpenHands Agent CanvasFree local; Teams $500/mo + $50/seat; Enterprise custom [57]Teams: 30-day trial, RBAC, SSO; no free hosted tier [71]$50 a month per extra seat [57]Teams: 10 users, 5,000 sessions a month [57]
Warp cloud agentsFree; Build $20/mo; Max $200/mo; Business $50/user/mo [62]Build $20 usage/mo; Max $200; Business $20/seat, 25 seats [72]$3/agent hour + inference + compute [73]; No compute charge on self-hosted compute [74]Included usage stops at 25 Business seats [72]

Prices are each vendor's list price on 2026-11-06. Model tokens are a separate line item wherever the agent runs on your own subscription or key.

What is a background coding agent?

A background coding agent is a coding agent you hand work to and walk away from. It does not share your editor or your laptop. Instead it runs in a cloud environment that has your repository, dependencies and tools in place, so a task keeps going after the lid closes. [3] [13] The request arrives from somewhere a team already lives: a Slack thread, a web dashboard, a schedule, a webhook or a ticket. [4] [75] The output is something a reviewer can accept or reject, most often a pull request, sometimes a branch or a comment. [5] [20]

Three things separate the category from editor assistants. Asynchrony: nobody watches it type. Isolation: it works on its own machine or sandbox, not yours. And a review gate: the agent proposes, a human merges. Vendors use other names, including cloud agents, autonomous agents and Droids, but the shape is the same. [76] [77]

How we ranked them

We scored each agent on the five columns in the first table. Whose model bill it lands on, and whether a subscription you already hold counts. Where the work runs, and how much of that you control. Which entry points exist, with Slack weighted heavily because that is where small teams actually ask for things. How finished work ships, favouring a reviewable pull request over a branch you must publish yourself. And price, in units a finance team can check.

The reader we optimised for is an engineering team between five and twenty strong, with Claude or ChatGPT seats already on the invoice, a backlog nobody reaches, and no one whose job is running an agent platform. A different buyer gets a different order, so every entry below names the buyer for whom that agent would rank first.

Disclosure: Nori publishes this list and makes Jiro, which it hosts in its own cloud and, for customers who want it, as a dedicated deployment in their own AWS account. [78] [79] We cite every rival claim to the vendor's own page and give each vendor's strongest case. Discount our first place accordingly. Charlie Labs is absent because it shut down on 2026-10-05. [80]

The ranking

Twelve agents, best fit for our reader first. Prices and limits are dated 2026-11-06 and linked in the sources list.

1. Jiro: first for small teams with Claude or ChatGPT seats already paid for

Jiro's decisive fact is the bill. Sign Claude Code in with the Anthropic key, Claude login or setup token you already use; Codex takes Codex OAuth or an OpenAI key; and Nori adds nothing to the token price. [1] [2] You choose the harness from five first-class agents (Codex, Claude Code, OpenCode, pi-acp and Cursor Agent, the last needing a Cursor key). [81] [1]

Each session gets an isolated cloud environment, prepared by your organisation's setup script, with one machine serving one active session. [3] [82] [83] Mention the bot in Slack and the whole thread becomes the session; Discord, the dashboard on the web, a terminal, cron jobs and incoming webhooks all reach the same fleet. [84] [85] What returns is reviewable: a PR, a commit or a comment. [5] GitHub access is a GitHub App scoped to selected repos, or a PAT. [86]

Pricing: the trial is 30 days and asks for no card; Developer is $99 monthly (one user, three runtimes); Team is $250 monthly (several users, five runtimes shared). [6] Runtimes sleep when idle. [87] Miravoice credits Nori with over a quarter of the PRs it merges. [88] Every session is logged and searchable. [89]

2. Cursor cloud agents: first for teams already on Cursor Premium

Cursor's cloud agents, the renamed background agents, return merge-ready pull requests: each comes with screenshots, video and logs, and the agent opens one PR per repository it touches. [76] [10] You can kick off a run from Slack, from a GitHub or Bitbucket comment, from Linear, the API, the desktop app, the web or the iOS and Android apps, against repositories hosted on GitHub, GitLab, Azure DevOps or Bitbucket Cloud. [9] Runs happen in Cursor's cloud VMs or, if you prefer, on machines you host yourself. [8]

The catch is the meter. Cloud runs draw on your plan's usage at the selected model's rate, and the Premium tier, $30 a month through the X Subscription, is the floor; models from third parties are invoiced separately, at list API rates. [7] [11] Anyone still on legacy Pro, Pro+ or Teams may stay there. [90] If your engineers already code in Cursor, this ranks first. If they do not, you are buying a second model bill.

3. OpenAI Codex cloud: first for ChatGPT-only teams spending nothing extra

Codex cloud is the cheapest way onto this list for a ChatGPT shop. A $20 Plus seat already includes it; Pro is sold at $100, $200 or $500 a month, Business at $25 per user monthly or $20 on an annual contract, and the usage ceiling is whatever your ChatGPT plan already has. [16] Tasks run on cloud VMs, so a sleeping laptop does not stop them. [13] A task can begin in ChatGPT itself, whether web, desktop or mobile, or at the Codex CLI; in Slack, @ChatGPT has replaced @Codex as the handle to use. [14] [91]

Two limits. OpenAI models only, and cloud features stay locked behind a ChatGPT plan; an API key alone opens the CLI, SDK and IDE, not the cloud. [12] And the agent stops short of the PR: the diff comes back for review and the commit, or the PR, is your move. [15] Separately, OpenAI intends to retire the legacy cloud behind its Linear and GitHub hooks, so check that roadmap before wiring tickets to it. [92]

4. Claude Code on the web: first for a solo engineer on Claude Pro or Max

Anthropic's own hosted sessions ride along with every paid plan: Pro ($20 a month), the single Max tier ($100), Ultra ($200), plus Team and Enterprise seats; Free has no Claude Code. [21] Max is one tier now, five times Pro's usage, with the old 20x tier sold as Ultra. [63] A session gets a fresh VM that Anthropic manages, sized at 4 vCPUs with 16 GB of memory and a 30 GB disk, unless your organisation routes sessions to its own self-hosted environment. [18]

Three ways in: claude.ai/code, the Claude apps on desktop and mobile, or the claude --cloud flag in a terminal, with claude --teleport bringing a cloud session home to your laptop. [19] The limitations for a team are structural rather than technical. Sessions are one person's, not a channel's; Slack work is routed to Claude Tag on Team and Enterprise. [93] The agent pushes a branch; the pull request is yours to open, and from there auto-fix can push fixes for red CI or reviewer comments. [20] Models are Claude only, through an Anthropic account. [17]

5. Devin: first for teams that want one vendor to own agent and model

Devin keeps working after it opens the PR: it replies to reviewer comments until someone archives the session. [25] Automations can kick off sessions on a schedule, from a webhook, or from events in PagerDuty, Pylon, Jira, Linear, GitLab, GitHub and Slack. [75] Repos on GitHub, GitLab, Bitbucket and Azure DevOps all work, and so do Microsoft Teams requests. [24] Model choice covers OpenAI, Google and Anthropic plus Cognition's own SWE line, and an Adaptive setting routes each task automatically. [22]

Self-serve pricing changed on 2026-04-14 to quota plans billed in dollars. [94] Today that means a Free tier, Pro ($20 a month), Max ($200) and Teams ($40 a seat against an $80 floor each month, which prices one seat and two identically); spend the quota and you top up with on-demand credits. [26] Free and Pro also limit you to 10 simultaneous sessions. [65] Enterprise can place the Devbox in a dedicated VPC, but the brain stays in Cognition's cloud. [95] Your existing Claude or ChatGPT plan counts for nothing here, which is why Devin sits fifth for our reader.

6. Copilot in the cloud: first for GitHub-native teams with a procurement gate

GitHub now calls the Copilot coding agent Copilot in the cloud. It researches the repo, plans and implements inside GitHub's cloud sandbox, on a GitHub Actions runner, or in a remote environment of your own. [96] Actions runs are capped at 59 minutes, which rules out long jobs. [28] The surface list is the longest on the page: GitHub.com and Mobile; the VS Code, JetBrains, Visual Studio and Eclipse IDEs; the GitHub CLI, its REST API and its MCP server; Raycast; and Slack, Teams, Jira, Linear and Azure Boards. [29]

Output is a draft PR on a branch of its own, and approving or merging that PR is beyond its permissions, a conservative default many security teams like. [30] An individual pays $10 monthly on Pro, $39 on Pro+ or $100 on Max. Business seats cost $19 per user and Enterprise seats $39, while Free excludes the cloud agent entirely. [31] All plans have billed through GitHub AI Credits since 2026-06-01, metering tokens at per-model rates with a credit worth $0.01. [66] GitHub supplies the models (Claude, GPT, Gemini and MAI); own keys are not documented for the cloud agent. [27] Paid plans can also assign tasks to Claude or Codex as third-party agents, in public preview. [97]

7. Claude Tag: first for Claude Team or Enterprise orgs that want Anthropic alone

Claude Tag is a shared @Claude in Slack channels: anyone tags it, it replies in the thread, and the channel watches. [98] Anthropic announced it on 2026-06-23 and still labels it a public beta. [99] [100] Each thread gets a throwaway sandbox, built on the engine behind Claude Code on the web, or runs on self-hosted runners without admin connectors. [34] [101] Code goes through the Claude GitHub App as draft pull requests authored by Claude. [36]

The constraints are the plan and the harness. It is Team and Enterprise only, not Pro, Max or Ultra. [67] Channel work bills by usage from an organisation balance, and on Team the bot stays silent until credits are bought, though owners can set org and per-channel spend caps. [33] [37] [68] Repository hooks and .mcp.json go unread, the effort level cannot be changed, and every session runs in auto mode. [102] Routines run on schedules, but not on repo events like every new pull request. [103] Zero Data Retention orgs are excluded. [104]

8. Factory: first for larger organisations that want a whole platform

Factory sells agent-native software development, and its agents are Droids. [77] A Droid can run on your laptop via the CLI or desktop app, on Factory's hosted Droid Computers, on hardware you own (Remote Control), or in your cloud account (OpenCompute). [39] Slack, Microsoft Teams, Jira and Linear can all delegate work, and a delegated Linear issue returns a pull request URL. [40] [41] Starter automations (Ticket to PR, Slack Bug Fixer, Alert Responder) fire from schedules or events and open PRs. [105]

Bring-your-own keys and local models are CLI and desktop features; neither the hosted web app nor the mobile app accepts them, which matters for anyone buying this for background work. [38] Solo plans: Pro $20, Plus $100, Max $200, all monthly. Teams: $60 a month plus $40 per seat, maximum 10 seats. [42] Factory Private, announced 2026-09-18, is still in private preview. [106] A lot of product. More than a ten-person team will configure.

9. Tembo: first for teams that need Sentry and PagerDuty triggers on their own keys

Tembo is the closest match to Jiro's model: several harnesses, your own credentials. Each task can use a different harness (Pi, OpenCode, Codex, Claude Code), and you pay the model side yourself through Anthropic, OpenAI or xAI keys or a ChatGPT, Claude or SuperGrok plan. [43] Its homepage adds Cursor and Amp to the runtime list. [107] The trigger list is the widest of the own-key agents: schedules and webhooks, Sentry and PagerDuty alerts, Jira and Linear tickets, Slack messages, and events from GitHub, GitLab or Bitbucket. [45] Issues and alerts become pull requests, including coordinated PRs across several repos. [46]

Sessions run in Docker or VM environments with a terminal and a live desktop. [44] Pro costs $60 a month and bundles $90 of usage; Max costs $200 and bundles $400; VM time is metered per second, the smallest size at $0.1066 an hour. [47] Pro covers up to 5 users and Max up to 10; the only free option is a 7-day trial. [69] Busy months cost more than the sticker. If incident response is your main use, Tembo ranks first.

10. Amp: first for trunk-based solo developers who want the cheapest start

Amp's default, called Ship, rebases, then tests, then lands the result directly on the base branch with no pull request; enabling Push to Branch changes that to a feature branch plus a PR link. [51] Each thread occupies an orb, a remote machine that Amp hosts or that you host through your own provider; runners on your own hardware are the alternative. [49] It accepts ChatGPT, Claude and other subscriptions and API keys with no token fee outside Enterprise, with Claude plans joining on 2026-10-10. [48] [108]

The Hobby tier costs nothing beyond pay-as-you-go orb time; Megawatt ($20 a month) includes $20 of agent usage and 750 hours on a1.small orbs; Gigawatt ($200) includes $200 and 1,000 hours on a1.xxlarge. [52] Orb time is charged by the minute, starting at $0.08 an hour. [70] You reach it from ampcode.com, native apps on macOS, iOS and Android, a CLI and SDKs, or through GitHub and Slack; the Slack AI agent app requires a paid Slack workspace. [50] [109] Its VS Code and Cursor extensions went dark on 2026-03-05. [110] Cautious teams should flip the Ship default before inviting it anywhere.

11. OpenHands Agent Canvas: first for teams that must self-host and audit the source

OpenHands has repositioned. Its README now presents the project as Agent Canvas, a self-hosted, always-on control center for coding agents and automations, under an MIT licence. [111] [112] The default install is local; agents can also live in Docker containers, on VMs, or on OpenHands' own Cloud or Enterprise hosting if you opt in. [54] The harness can be its native agent or Pi, Gemini, OpenCode, Codex or Claude Code, against any LLM. [53] Automations fire on schedules or webhooks. [113]

The old free hosted tier is gone from the pricing page. Self-hosting Agent Canvas costs nothing for a single user; Teams runs $500 a month, covers 10 users, adds seats at $50 a month each, caps usage at 5,000 sessions a month, and comes with a 30-day trial, RBAC and SSO. [57] [71] On GitHub, an issue carrying the openhands label, or a comment that starts with @openhands, triggers a run; if the run resolves the issue it opens a PR. [56] The cloud UI, API and Slack app still exist. [55] The default install runs on your own machine, so your team operates the host unless you pay for OpenHands Cloud or Enterprise hosting. [54]

12. Warp cloud agents: first for teams that type in Warp and want to hand a local session to the cloud

Warp makes the ranked list this time because its docs now say what a cloud run produces: a branch, a pull request, a report or a transcript, with a warning that a completed run is no guarantee the result is correct. [61] Its signature trick is the handoff. Press & or type /handoff mid-conversation in the local Warp Agent and the transcript forks into a cloud run, taking your uncommitted edits and untracked files along. [114] Beyond that, runs begin from Slack, Linear, Jira, GitHub, schedules, the Oz CLI, the API or the web app, and oz-agent-action launches them from pull request events, issue labels, daily schedules or red CI inside your own workflows. [60] [115]

Twelfth because of keys and meters. Your personal Anthropic or OpenAI key stays on the laptop; cloud runs get team-managed inference only on Business or Enterprise. [58] Each run costs $3 per agent hour on top of inference and compute, though compute is free when the sandbox sits on hardware you host. [73] [74] Build ($20 monthly) carries $20 of usage, Max ($200) carries $200, and Business ($50 per user) carries $20 a seat, up to 25 seats. [62] [72] Pick Warp first if your engineers already live in its terminal.

Who should not pick Jiro

Jiro is the wrong first choice in four situations, and we would rather say so than win a trial you will cancel.

You have no Claude or ChatGPT spend and no wish to start any. Jiro's value is putting credentials you already hold to work; without them, a bundled product such as Devin or Copilot gives you agent and model in one invoice. [1] [26] [31]

Your tickets live in Linear or Jira and you want them to start work by themselves. Sourced entry points for Jiro are Slack and Discord, the web dashboard, a terminal, cron schedules and webhooks; Tembo, Devin and Copilot listen to Linear and Jira events directly. [4] [45] [75]

You need Gemini. Nori's first-class agents are five: pi-acp, OpenCode, Cursor Agent, Codex and Claude Code, and Gemini was removed from managed Sessions. [81] [116]

You want to run the control plane yourself. Nori hosts Jiro; the private alternative is a dedicated deployment on EC2 inside AWS infrastructure you control; OpenHands Agent Canvas is the self-host-everything pick. [78] [111]

Also considered: Kiro and Jules

Two names appear on most 2026 lists and are not ranked here.

Kiro is Amazon's spec-driven agent; Kiro Web launched on 2026-05-18 for paid subscribers, with an Autonomous toggle under which it plans, writes, verifies and returns a pull request. [117] [118] Up to 10 cloud sessions at a time run in Amazon's managed sandbox, and PRs come from a kiro label or /kiro comment. [119] [120] Plans run from Free with 50 credits to Power at $200 a month, and Kiro Web needs a paid plan. [121] It is unranked for two reasons: its FAQ lists an IDE, a CLI, a browser, a phone app and Kiro Crew as entry points, with no chat surface, and every model is metered in Kiro credits rather than on a Claude or ChatGPT plan you already hold. [122] [123]

Jules is being turned off. Google's changelog entry of 2026-10-29 sets 2026-11-11 as the last day to finish sessions; from then Jules is read-only, and it closes for good on 2026-12-31. [124] Do not start a pilot on it. Before the deadline, the Jules Tools CLI plus an export prompt from the notice can pull out sessions, scheduled tasks, repositories, memories and setup scripts; code already pushed stays in GitHub. [125] Readers who want to stay inside Google should look at Google Antigravity, which the same notice names as the agent-first platform that Jules's learnings now feed. [126] Antigravity's own site calls it agentic, built around an agent manager, and free for developers. [127] Jules was Gemini-only and aimed at personal accounts anyway. [128] [129]

Whose model bill does each agent land on?

Money decides more than features. Anthropic's enterprise data puts Claude Code at roughly $13 for each active developer-day, which it translates to $150 to $250 monthly, so a ten-person team is already spending real money on tokens before any agent platform is added. [130] The question is whether a second product resells those tokens.

Three billing shapes exist. Own credentials with no markup: Jiro, Tembo and Amp accept your Claude or ChatGPT plan or API key and charge only for the platform or compute. [2] [43] [48] Bundled model access: Devin, Copilot and Factory's hosted apps sell quota or credits that include the model. [26] [66] [38] Vendor-locked plans: Codex cloud insists on a ChatGPT plan, while an Anthropic account is the ticket for Claude Code on the web, and Kiro only spends its own credits. [12] [17] [123]

Watch the metered cases. Cursor charges list API price for third-party models, Warp's pay-as-you-go runs 20% above API, and Copilot credits are $0.01 each at per-model rates. [11] [62] [66]

Where does the agent run, and does it matter?

It matters for two reasons: how long a task can run, and who can see the code. Copilot's Actions environment stops at 59 minutes, which is fine for a lint fix and useless for a migration. [28] Claude Tag threads live in an ephemeral sandbox that disappears afterwards. [34] Jiro gives each session an isolated cloud environment prepared by your setup script, with one machine per active session, and sizes plans by persistent runtimes that sleep when idle. [3] [83] [87]

On data boundaries, the options differ sharply. Jiro's hosted service is single-tenant per organisation, and the private option puts the session machines on EC2 in AWS infrastructure you control. [79] [78] Devin Enterprise places the Devbox in a dedicated VPC while the model side stays with Cognition. [95] Cursor can push tool execution to self-hosted machines; Amp's orbs can sit on a provider you host, or you can supply runners; Factory's OpenCompute uses your cloud. [8] [49] [39] Agent Canvas is self-hosted by definition. [111]

Which agents open the pull request themselves?

The review gate is where products quietly diverge, and it changes how much babysitting a team does. From Jiro you get a PR, commit or comment back, which the team then merges or bounces. [5] Cursor opens merge-ready PRs with screenshots and video. [10] Devin opens PRs and keeps replying to review comments. [25] Copilot and Claude Tag both open draft pull requests, and Copilot cannot approve or merge its own. [30] [36] Tembo and OpenHands open PRs from issues and alerts. [46] [56]

Two products stop one step short. Codex cloud hands you the diff and you commit or open the PR. [15] Claude Code on the web pushes a branch and leaves the PR step to you, with auto-fix following up on failing CI and review notes. [20] One goes a step too far for most teams: Amp's default Ship lands on the base branch once tests pass, unless you switch to Push to Branch. [51] Pick the gate that matches your branch protection, not the demo.

Which background coding agents work from Slack?

Slack is the surface that makes a background agent a team tool rather than a personal one, and the implementations differ. Jiro is mentioned in a channel and replies in a thread, and the thread itself is the session: each reply continues it, with the same fleet behind Discord, the web and the terminal. [84] [4] Claude Tag is Slack-native too, for Team and Enterprise customers, and Anthropic lists Microsoft Teams as coming soon. [35] Anthropic's earlier Slack bot was switched off on 2026-10-05. [131]

Cursor, Devin, Copilot, Factory, Tembo, Warp and Amp all take Slack requests, with Amp's AI agent app needing a paid Slack plan. [9] [24] [29] [40] [45] [60] [109] Codex requests in Slack go to @ChatGPT. [91] Claude Code on the web has no team Slack surface of its own; Anthropic points Team and Enterprise to Claude Tag. [93]

For a team on Claude Pro or Max rather than Team, the practical route to Claude Code in Slack is a subscription token. Run claude setup-token once; what comes out is a twelve-month token bound to the paid Claude plan on that account, anything from Pro to Enterprise, and Jiro accepts it as a credential. [132] [1]

What changed in this category during 2026?

A lot moved, and older lists are wrong in specific ways. Charlie Labs closed on 2026-10-05. [80] Cognition replaced Devin's Core and Team plans with quota-based self-serve plans on 2026-04-14 and relaunched Windsurf as Devin Desktop on 2026-06-02. [94] [133] GitHub moved every Copilot plan to AI Credits on 2026-06-01 and now markets the coding agent as Copilot in the cloud. [66] [96]

Anthropic launched Claude Tag on 2026-06-23, retired the old Slack bot on 2026-10-05, and collapsed Max into one $100 tier with Ultra at $200. [99] [131] [63] Cursor's background agents became Cloud Agents, sold through the X Subscription with Premium at $30. [76] [11] Amp killed its editor extensions on 2026-03-05, dropped ads on 2026-03-30 and added Claude plans on 2026-10-10. [110] [134] [108] OpenHands became Agent Canvas, with Teams at $500 a month. [111] [57] Kiro Web launched for paid plans on 2026-05-18. [118] Google gave Jules a 2026-12-31 end date. [124]

How should a small team run a pilot?

Pick two finalists and give both the same boring work for two weeks: a flaky test, a stale README and an alert from last Tuesday. Start each run from where your team would really start it, which for most small teams is a Slack thread. [84]

Measure three numbers. First, what fraction of returned pull requests a reviewer would merge without rewriting; Miravoice reports Nori behind more than 25% of its merged PRs, which is the kind of figure you want to reproduce rather than take on trust. [88] Second, the real monthly cost including tokens, compute and overage, using the units in the pricing table. [47] [26] Third, how often someone had to open a laptop to finish the agent's job.

Then set the review gate deliberately. Keep draft or review-only defaults where they exist, and switch Amp off Ship before the pilot starts. [30] [51] Use the trial period rather than paying upfront: Jiro's is 30 days with no card, OpenHands Teams has 30 days, Tembo 7 days. [6] [71] [69]

Jiro at a glance

FactJiro
What it isA background coding agent: it works in the cloud and returns pull requests for review [5]
Made byNori (Tilework Inc.), founded by Amol Kapoor, ex-Google Research [135]
Agents it runsClaude Code, Codex, Cursor, OpenCode, pi-acp [81]
Whose model billYour Claude/ChatGPT login or API key; Cursor key [1]; Your keys, zero token markup [2]
PriceFree 30-day trial; Developer $99/mo; Team $250/mo [6]
RuntimesDeveloper: up to 3; Team: 5 shared; trial: 5 for 30 days [6] [87]
Where work startsSlack, web, Discord, terminal, schedules, webhooks [4]
Where it runsIsolated cloud environment with your repos and tools [3]; Single-tenant per organization [79]
ReposGitHub App (all or selected repos) or a PAT [86]
Not to be confused withAmazon's Kiro (IDE and agent) or Atlassian's Jira (issue tracker)
Last verified2026-11-06

About the publisher

Jiro is made by Nori (Tilework Inc.), a New York company. Nori was founded by Amol Kapoor, its CEO, who before Nori did AI research at Google on graph neural networks and co-authored MixHop (ICML 2019). [135] [136] Clifford Ressel is Nori's co-founder and CTO. [137] Nori's customers include Corvus and Miravoice. [138] [88]

FAQ

What is a background coding agent?

A coding agent that works asynchronously in a cloud environment with your repository, started from chat, a schedule or a ticket, and returns a pull request, branch or comment for a human to review. [3] [5]

Which background coding agent is best for a small team in 2026?

For an engineering team of 5 to 20 that is already paying Anthropic or OpenAI for seats, Jiro, because it runs Claude Code or Codex on those credentials without markup. [1] [2] Cursor shops should pick Cursor cloud agents, and ChatGPT-only teams Codex cloud. [11] [16]

Is there a free background coding agent?

Self-hosted Agent Canvas is free for a single user, Amp's Hobby tier costs nothing beyond orb time, and Jiro offers a 30-day trial with no card. [57] [52] [6]

Can I run Claude Code in Slack on my own Claude subscription?

Yes, with Jiro: run claude setup-token to mint a twelve-month subscription token, then connect it as the Claude Code credential. [132] [1] Claude Tag also works in Slack but only on Team and Enterprise plans. [67]

Which background coding agents merge code without a review?

Only Amp by default, whose Ship pushes to the base branch after tests. [51] Copilot opens drafts it cannot merge, and Jiro returns pull requests for your team to review. [30] [5]

What happened to Charlie Labs?

Charlie Labs announced it was closing, and Charlie stayed available only until 2026-10-05. [80]

Does OpenHands still have a free hosted tier?

No. The pricing page lists Agent Canvas as free to self-host, Teams at $500 a month covering 10 users, and Enterprise as custom; no free cloud tier appears. [57] [71]

How much does Devin cost for a team?

Devin Teams charges $40 per seat each month but bills at least $80, meaning a single seat costs $80, as does a pair; overage is billed as on-demand credits. [26]

Try Jiro

Start the 30-day free trial

No card needed. Or install Jiro's engineering skills into your own agent: npx nori-skillsets install public/jiro

Sources

  1. Claude Code in Nori accepts an Anthropic API key, a Claude OAuth login or a setup token; Codex accepts an OpenAI API key or Codex OAuth; Cursor Agent needs a Cursor API key; OpenCode and pi-acp take a compatible API base URL, API key and model. https://docs.noriagentic.com/sessions/reference/supported-agents-and-compute (checked 2026-11-06)
  2. Nori's comparison table lists any model on your own keys with flat pricing and zero token markup. https://noriagentic.com/#compare (checked 2026-11-06)
  3. Each Nori agent works in an isolated cloud environment with your repository, tools, dependencies and services ready to use. https://noriagentic.com/#faq (checked 2026-11-06)
  4. Nori work can start from Slack, the web dashboard, Discord, the Handroll terminal, schedules and webhooks, all reaching the same fleet. https://docs.noriagentic.com/start/what-can-nori-do#start-agent-work-from-the-tools-your-team-already-uses (checked 2026-11-06)
  5. Nori agents can return a pull request, commit, comment or completed task for you to review, merge or send back. https://noriagentic.com/#faq (checked 2026-11-06)
  6. Nori's plans start with a free 30-day trial with no card; Developer is $99 a month for one user with up to three persistent runtimes, and Team is $250 a month for multiple users sharing five runtimes. https://noriagentic.com/#pricing (checked 2026-11-06)
  7. Cloud agents use your plan's usage like other agent work, at the selected model's rate. https://cursor.com/docs/cloud-agent (checked 2026-11-06)
  8. Cursor cloud agents run in isolated cloud VMs with full development environments, or move tool execution to self-hosted machines you manage. https://cursor.com/docs/cloud-agent (checked 2026-11-06)
  9. Cloud agents start from the Cursor desktop app, the web, iOS and Android, Slack, GitHub and Bitbucket comments, Linear and an API, on GitHub, GitLab, Bitbucket Cloud or Azure DevOps repos. https://cursor.com/docs/cloud-agent (checked 2026-11-06)
  10. Cursor cloud agents produce merge-ready PRs with screenshots, videos and logs, and open a PR in each repo they touch. https://cursor.com/docs/cloud-agent (checked 2026-11-06)
  11. Cursor subscriptions are sold only through X Subscription Corp; Premium at $30 a month is the first tier that includes cloud agents and automations, and third-party models are billed separately at API list price. https://cursor.com/pricing (checked 2026-11-06)
  12. Codex cloud runs OpenAI models; an API key unlocks Codex in the CLI, SDK and IDE but not cloud features, which need a ChatGPT plan. https://learn.chatgpt.com/docs/pricing.md (checked 2026-11-06)
  13. Codex cloud runs each task in a cloud VM, so work continues while your computer is asleep. https://learn.chatgpt.com/docs/environments/cloud-environments.md (checked 2026-11-06)
  14. Codex cloud tasks start from ChatGPT on the web, mobile and desktop, or from the Codex CLI. https://learn.chatgpt.com/docs/cloud.md (checked 2026-11-06)
  15. In Codex cloud you review the changes, then commit or open a pull request when you are ready. https://learn.chatgpt.com/docs/cloud.md (checked 2026-11-06)
  16. Codex cloud is included from ChatGPT Plus at $20 a month; Pro costs $100, $200 or $500 a month and Business $25 per user a month billed monthly or $20 billed annually, with usage limits shared with ChatGPT. https://learn.chatgpt.com/docs/pricing.md (checked 2026-11-06)
  17. Cloud sessions run Claude models and require an Anthropic account; they are not available through Bedrock, Google Cloud or other third-party providers. https://code.claude.com/docs/en/claude-code-on-the-web (checked 2026-11-06)
  18. A Claude Code cloud session runs in a fresh Anthropic-managed VM with 4 vCPUs, 16 GB of RAM and 30 GB of disk, or in an organization's self-hosted environment when routed there. https://code.claude.com/docs/en/cloud-environments (checked 2026-11-06)
  19. Cloud sessions start from claude.ai/code, the Claude mobile and desktop apps, and claude --cloud in a terminal, and claude --teleport pulls a session back to your machine. https://code.claude.com/docs/en/claude-code-on-the-web (checked 2026-11-06)
  20. Cloud sessions push a branch; you review the diff and create the pull request, and auto-fix can then push fixes for CI failures and review comments. https://code.claude.com/docs/en/claude-code-on-the-web (checked 2026-11-06)
  21. Claude Code is included in Claude Pro ($20 a month, or $17 a month billed annually), Max ($100 a month), Ultra ($200 a month) and Team and Enterprise seats; the Free plan has no Claude Code. https://claude.com/pricing (checked 2026-11-06)
  22. Devin offers models from Anthropic, OpenAI, Google and Cognition's own SWE family, with an Adaptive option that routes each task to a model automatically. https://docs.devin.ai/cli/models.md (checked 2026-11-06)
  23. Devin Cloud runs sessions on Cognition-hosted machines; Devin CLI and Devin Desktop run locally. https://docs.devin.ai/llms.txt (checked 2026-11-06)
  24. Devin connects to GitHub, GitLab, Bitbucket and Azure DevOps repos and takes work from Slack, Microsoft Teams, Linear, Jira, PagerDuty and Pylon. https://docs.devin.ai/integrations/overview.md (checked 2026-11-06)
  25. Devin opens pull requests and automatically responds to comments on PRs its sessions are working on until the session is archived. https://docs.devin.ai/integrations/gh.md (checked 2026-11-06)
  26. Devin's self-serve plans are Free, Pro at $20 a month, Max at $200 a month and Teams at $40 a month per full seat with an $80 monthly minimum, so one seat and two seats both cost $80 a month; usage past the quota is billed as on-demand credits. https://docs.devin.ai/admin/billing/self-serve.md (checked 2026-11-06)
  27. The cloud agent offers Auto plus GitHub-provided Claude, Gemini, GPT and MAI models; bring-your-own-key is documented for IDEs, Copilot CLI, the Copilot app and the SDK, not for the cloud agent. https://docs.github.com/en/copilot/reference/ai-models/supported-models (checked 2026-11-06)
  28. Copilot cloud agent sessions run in a GitHub cloud sandbox, in a GitHub Actions environment capped at 59 minutes, or in a remote environment you connect. https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent (checked 2026-11-06)
  29. Copilot cloud agent sessions start from GitHub.com, GitHub Mobile, VS Code, JetBrains, Visual Studio and Eclipse, the GitHub CLI, REST API and MCP server, Raycast, Slack, Microsoft Teams, Linear, Jira and Azure Boards. https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/create-a-pr (checked 2026-11-06)
  30. Copilot cloud agent opens draft pull requests on its own branch and cannot approve or merge them. https://docs.github.com/en/copilot/concepts/agents/cloud-agent/risks-and-mitigations (checked 2026-11-06)
  31. Copilot Pro is $10, Pro+ $39 and Max $100 a month; Business is $19 and Enterprise $39 a month per user; Copilot Free does not include the cloud agent. https://docs.github.com/en/copilot/get-started/plans (checked 2026-11-06)
  32. Claude Tag runs Anthropic's Opus and Sonnet models, chosen per thread, per channel or by admin default. https://claude.com/docs/claude-tag/users/models (checked 2026-11-06)
  33. Claude Tag channel work is billed by usage from an organization balance rather than per seat; one-to-one DMs bill to the sender's own seat. https://claude.com/docs/claude-tag/overview (checked 2026-11-06)
  34. Each Claude Tag thread runs in its own ephemeral sandbox in Anthropic's cloud, on the same engine as Claude Code on the web. https://claude.com/docs/claude-tag/concepts/how-it-works (checked 2026-11-06)
  35. Claude Tag works in Slack channels and threads for Team and Enterprise customers, and Anthropic lists Microsoft Teams as coming soon. https://claude.com/product/tag (checked 2026-11-06)
  36. Claude Tag code work goes through the Claude GitHub App, which pushes branches and opens draft pull requests authored by Claude; GitLab works through a service-account token. https://claude.com/docs/claude-tag/admins/configure-github (checked 2026-11-06)
  37. On the Team plan, Claude won't respond in channels until the organization buys usage credits. https://claude.com/docs/claude-tag/admins/set-spend-limit (checked 2026-11-06)
  38. Factory routes across Anthropic, OpenAI and other models and accepts your own keys or local models in the CLI and desktop app, but not in the hosted web or mobile apps. https://docs.factory.com/model-independence/byok.md (checked 2026-11-06)
  39. Droids run locally in the Droid CLI and Factory App, on Factory-hosted Droid Computers, on your own computer via Remote Control, or in your own cloud via OpenCompute. https://docs.factory.com/llms.txt (checked 2026-11-06)
  40. Droids work in the Factory App, Droid CLI, headless Droid Exec and IDEs, and take delegations from Slack, Linear, Jira and Microsoft Teams. https://docs.factory.com/llms.txt (checked 2026-11-06)
  41. When delegated a Linear issue, a Droid is instructed to open a pull request and return its URL. https://docs.factory.com/delegations/linear.md (checked 2026-11-06)
  42. Factory's plans are Pro at $20 a month, Plus at $100, Max at $200, and Teams at $60 a month plus $40 per seat for up to 10 seats; Business and Enterprise are custom. https://factory.com/pricing (checked 2026-11-06)
  43. Tembo runs Claude Code, Codex, OpenCode or Pi per task and accepts your own OpenAI, Anthropic or xAI keys or a Claude, ChatGPT or SuperGrok subscription. https://docs.tembo.io/features/models.md (checked 2026-11-06)
  44. Tembo runs coding agents in isolated Docker or VM sessions with code, terminal, files and a live desktop. https://www.tembo.io/ (checked 2026-11-06)
  45. Tembo starts work from GitHub, GitLab, Bitbucket, Slack, Linear, Jira, Sentry, PagerDuty, webhooks and schedules. https://www.tembo.io/ (checked 2026-11-06)
  46. Tembo turns issues and alerts from Sentry, Linear and Jira into pull requests, including coordinated multi-repo PRs across GitHub, GitLab and Bitbucket, and reviewer feedback can be sent back into the same session. https://www.tembo.io/ (checked 2026-11-06)
  47. Tembo's Pro plan is $60 a month with $90 of included usage and Max is $200 a month with $400, with cloud VMs metered by the second from $0.1066 an hour. https://www.tembo.io/pricing (checked 2026-11-06)
  48. Amp accepts ChatGPT, Claude and other subscriptions as well as API keys, and charges no token fees on your own keys outside its Enterprise tier. https://ampcode.com/news/free-agent (checked 2026-11-06)
  49. Amp runs cloud work in orbs, one remote machine per thread on Amp-hosted or self-hosted providers that keeps working while your laptop is closed, or on your own machines as runners. https://ampcode.com/manual/orbs (checked 2026-11-06)
  50. Amp runs from the web app at ampcode.com, macOS, iOS and Android apps, the CLI, SDKs, and GitHub and Slack integrations. https://ampcode.com/manual (checked 2026-11-06)
  51. Amp's default Ship rebases, runs tests and pushes straight to the base branch; the Push to Branch setting pushes a feature branch and reports a pull request URL instead. https://ampcode.com/docs/markdown/orbs/shipping (checked 2026-11-06)
  52. Amp's Hobby tier is free with pay-as-you-go orbs or your own runners; Megawatt is $20 a month with $20 of agent usage and 750 hours of a1.small orbs, and Gigawatt is $200 a month with $200 of agent usage and 1,000 hours of a1.xxlarge orbs. https://ampcode.com/docs/markdown/pricing (checked 2026-11-06)
  53. OpenHands runs its own agent or Claude Code, Codex, Gemini, Pi, OpenCode or any ACP agent, and works with any LLM. https://github.com/OpenHands/OpenHands (checked 2026-11-06)
  54. OpenHands runs locally by default, with agents running locally, in Docker, on VMs, or optionally on OpenHands Cloud or OpenHands Enterprise infrastructure. https://github.com/OpenHands/OpenHands (checked 2026-11-06)
  55. OpenHands Cloud works from its web UI and API, from GitHub, GitLab and Bitbucket repos, and from Slack. https://docs.openhands.dev/openhands/usage/cloud/cloud-ui (checked 2026-11-06)
  56. Labelling a GitHub issue openhands or commenting @openhands starts a run that opens a PR if it resolves the issue. https://docs.openhands.dev/openhands/usage/cloud/github-installation (checked 2026-11-06)
  57. Agent Canvas is free to self-host for a single user; Teams costs $500 a month for up to 10 users with extra seats at $50 a month each and a cap of 5,000 sessions a month, and Enterprise is custom. https://openhands.dev/pricing (checked 2026-11-06)
  58. Each Warp cloud run can use Warp Agent, Claude Code or Codex; personal BYOK keys are not copied to cloud runs, but Business and Enterprise can configure team-managed inference. https://docs.warp.dev/platform/handoff/local-to-cloud/ (checked 2026-11-06)
  59. Warp cloud agents run on Warp-hosted compute or on your own infrastructure. https://docs.warp.dev/agent-platform/cloud-agents/overview (checked 2026-11-06)
  60. Warp cloud runs start from Slack, Linear, Jira, GitHub, GitHub Actions, schedules, the Oz CLI, the API or the web app. https://docs.warp.dev/platform/integrations/ (checked 2026-11-06)
  61. A Warp cloud run yields results such as a branch, pull request, report or transcript, and Warp warns that a completed run alone does not confirm the result is correct. https://docs.warp.dev/agent-platform/cloud-agents/overview (checked 2026-11-06)
  62. Warp's Terminal and Agents plans are Free, Build at $20 a month, Max at $200 a month and Business at $50 per user a month; Warp Factories also offers pay-as-you-go usage at 20% above API rates. https://www.warp.dev/pricing (checked 2026-11-06)
  63. Claude Max costs $100 a month for five times the Pro plan's usage every five hours and every week; the former Max 20x tier is now sold as Ultra. https://support.claude.com/en/articles/11049741-what-is-the-max-plan (checked 2026-11-06)
  64. Usage past a Pro, Max or Ultra plan's limits can continue as usage credits billed at standard API rates. https://support.claude.com/en/articles/12429409-extra-usage-for-paid-claude-plans (checked 2026-11-06)
  65. Free and Pro allow up to 10 concurrent Devin sessions; Max and Teams have no concurrency cap. https://devin.ai/pricing (checked 2026-11-06)
  66. Since 2026-06-01, all Copilot plans bill in GitHub AI Credits, which meter input, output and cached tokens at per-model rates, with one credit equal to $0.01. https://docs.github.com/en/copilot/concepts/billing-and-usage/individuals/billing (checked 2026-11-06)
  67. Claude Tag is available only on Team and Enterprise plans, not on Free, Pro, Max or Ultra, and not through third-party deployments. https://claude.com/docs/claude-tag/overview (checked 2026-11-06)
  68. Claude Tag owners can set a monthly organization spend limit and per-channel limits, and Claude posts a warning in Slack as usage nears each limit. https://claude.com/docs/claude-tag/admins/set-spend-limit (checked 2026-11-06)
  69. Tembo Pro covers up to 5 users and Max up to 10, both with overage billing past the included usage; the only free option is a 7-day trial of Max, and Enterprise is custom. https://www.tembo.io/pricing (checked 2026-11-06)
  70. Orbs are billed per minute, from $0.08 an hour for the smallest size. https://ampcode.com/docs/markdown/orbs/sizes-and-costs (checked 2026-11-06)
  71. The OpenHands Teams plan is sold with a 30-day trial and adds multi-user identity, RBAC, budgeting controls and SSO/SAML on top of the free self-hosted Agent Canvas, whose pricing page lists no free hosted tier. https://openhands.dev/pricing (checked 2026-11-06)
  72. Warp's Build plan includes $20 of agent usage each month at API rates, Max includes $200 a month, and Business includes $20 per seat a month for up to 25 seats. https://www.warp.dev/pricing (checked 2026-11-06)
  73. Warp bills a cloud run for agent time at $3 per agent hour, plus inference and Warp-hosted compute. https://docs.warp.dev/support-and-community/plans-and-billing/platform-credits/ (checked 2026-11-06)
  74. Warp charges compute only for sandboxes on Warp-hosted compute, so local or self-hosted compute incurs no compute charge, and customer-supplied inference is billed by your own provider. https://docs.warp.dev/support-and-community/plans-and-billing/platform-credits/ (checked 2026-11-06)
  75. Devin Automations start sessions from schedules, webhooks, and Slack, GitHub, GitLab, Linear, Jira, Pylon and PagerDuty events. https://docs.devin.ai/product-guides/automations.md (checked 2026-11-06)
  76. Cursor's help page on background agents says that in Cursor these are called Cloud Agents. https://cursor.com/help/ai-features/background-agents (checked 2026-11-06)
  77. Factory sells agent-native software development; its agents are called Droids. https://factory.com/ (checked 2026-11-06)
  78. Nori hosts Jiro in its own cloud, and a dedicated BYOC deployment runs private EC2 session machines inside customer-controlled AWS infrastructure. https://docs.noriagentic.com/sessions/reference/supported-agents-and-compute (checked 2026-11-06)
  79. Each hosted Nori organization gets a single-tenant broker on its own subdomain. https://docs.noriagentic.com/sessions/concepts/security-and-credentials#organization-isolation (checked 2026-11-06)
  80. Charlie Labs announced it was shutting down, with Charlie available only until 2026-10-05. https://charlielabs.ai/blog/charlie-is-shutting-down (checked 2026-11-06)
  81. Nori's default first-class Sessions agents are Claude Code, Codex, Cursor Agent, OpenCode and pi-acp. https://docs.noriagentic.com/sessions/reference/supported-agents-and-compute (checked 2026-11-06)
  82. A Nori organization can define a setup script that installs dependencies and prepares every session's environment. https://docs.noriagentic.com/start/what-can-nori-do#give-agents-a-real-working-environment (checked 2026-11-06)
  83. In Nori, one claimed machine serves one active session within the organization's fleet. https://docs.noriagentic.com/resources/faq#is-each-session-isolated (checked 2026-11-06)
  84. Mention Nori's bot in a Slack channel to start a session; it replies in a thread, and every reply in that thread continues the same session. https://docs.noriagentic.com/sessions/guides/slack (checked 2026-11-06)
  85. Nori cron triggers run recurring jobs, reminders run once at a future time, and webhooks react to external events. https://docs.noriagentic.com/resources/faq#can-nori-run-tasks-without-a-person-online (checked 2026-11-06)
  86. Nori supports two GitHub credential methods: a GitHub App you install and scope to all repositories or a selected set, or a saved personal access token. https://docs.noriagentic.com/sessions/guides/integrations#github (checked 2026-11-06)
  87. Nori sizes plans by runtimes, which sleep when idle and wake on demand; the trial runs five for 30 days. https://noriagentic.com/#faq (checked 2026-11-06)
  88. Nori handles more than 25% of Miravoice's merged pull requests, and Miravoice's CTO says it easily saves him five hours a week. https://noriagentic.com/newsletter/2026-09-18-miravoice-case-study.html (checked 2026-11-06)
  89. Nori logs every session and every action, and makes them searchable. https://noriagentic.com/for-security-leaders.html (checked 2026-11-06)
  90. Legacy Cursor plans keep working as they are: Pro at $20, Pro+ at $60 and Ultra at $200 a month, and Teams Standard at $40 or Teams Premium at $120 per user a month; switching to The X Subscription is optional and permanent. https://cursor.com/docs/account/pricing/legacy-cursor-plans (checked 2026-11-06)
  91. Codex requests in Slack now start with @ChatGPT instead of @Codex. https://learn.chatgpt.com/docs/third-party/slack.md (checked 2026-11-06)
  92. Codex Cloud (Legacy) still powers Code Review and the GitHub and Linear integrations, and OpenAI plans to deprecate it. https://learn.chatgpt.com/docs/environments/cloud-environments.md (checked 2026-11-06)
  93. For Team and Enterprise, Slack work moves to Claude Tag (public beta), and the legacy Claude Code Slack bot was retired on 2026-10-05 in workspaces connected to a Claude organization. https://code.claude.com/docs/en/slack (checked 2026-11-06)
  94. Cognition retired its Core and Team plans for quota-based self-serve plans on 2026-04-14, billing self-serve overage in dollars instead of ACUs while Enterprise stays on ACUs. https://cognition.com/blog/new-self-serve-plans-for-devin (checked 2026-11-06)
  95. Devin Enterprise's Customer Dedicated Deployment runs the Devbox in a single-tenant VPC, while Devin's brain always stays in Cognition's cloud. https://docs.devin.ai/enterprise/deployment/overview.md (checked 2026-11-06)
  96. GitHub's Copilot cloud agent, now presented as Copilot in the cloud, researches a repository, plans changes and implements them in a GitHub cloud sandbox, a GitHub Actions environment or a remote environment you connect. https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent (checked 2026-11-06)
  97. On Pro, Pro+, Max, Business and Enterprise, Claude and Codex can be assigned GitHub tasks as third-party agents (public preview). https://docs.github.com/en/copilot/concepts/agents/about-third-party-agents (checked 2026-11-06)
  98. Claude Tag lets anyone in a Slack channel tag @Claude to hand it work; Claude replies in the thread, where the channel can see it. https://claude.com/docs/claude-tag/overview (checked 2026-11-06)
  99. Anthropic announced Claude Tag on 2026-06-23 as a beta for Claude Enterprise and Team customers. https://www.anthropic.com/news/introducing-claude-tag (checked 2026-11-06)
  100. Anthropic labels Claude Tag a public beta whose features may change before general availability. https://claude.com/docs/claude-tag/admins/migrate-from-earlier (checked 2026-11-06)
  101. Claude Tag sessions can run on self-hosted runners inside your network, but admin-added connectors are unavailable there. https://claude.com/docs/claude-tag/concepts/security-and-data (checked 2026-11-06)
  102. In Claude Tag, repository hooks and .mcp.json are not loaded, effort level is not configurable, and sessions run in auto mode. https://claude.com/docs/claude-tag/concepts/for-claude-code-users (checked 2026-11-06)
  103. Claude Tag supports scheduled routines, channel watches and single-PR subscriptions, but not routines fired by other repository events such as every new pull request. https://claude.com/docs/claude-tag/users/proactivity (checked 2026-11-06)
  104. Claude Tag isn't available to organizations with Zero Data Retention enabled. https://claude.com/docs/claude-tag/concepts/security-and-data (checked 2026-11-06)
  105. Factory automation starters such as Ticket to PR, Slack Bug Fixer and Alert Responder run on schedules or events and open pull requests. https://docs.factory.com/software-factory/automations.md (checked 2026-11-06)
  106. Factory Private, announced 2026-09-18 and still in private preview, places some or all of Factory's infrastructure in the customer's VPC. https://factory.com/news/factory-private (checked 2026-11-06)
  107. Tembo's homepage lists Claude Code, Codex/GPT, Cursor, Amp, OpenCode and Pi among the agent runtimes you can choose per task, with optional automatic model selection. https://www.tembo.io/ (checked 2026-11-06)
  108. Amp added support for running on a Claude subscription through the Claude Agent SDK on 2026-10-10. https://ampcode.com/news/use-your-claude-plan (checked 2026-11-06)
  109. Amp's Slack integration launched on 2026-07-20; its Slack AI agent app needs a paid Slack plan, while free Slack workspaces can still @mention Amp in channels. https://ampcode.com/docs/markdown/puck/slack-integration (checked 2026-11-06)
  110. Amp announced in February 2026 that its VS Code and Cursor extensions would shut off on 2026-03-05, leaving the CLI. https://ampcode.com/news/the-coding-agent-is-dead (checked 2026-11-06)
  111. OpenHands' README now presents the project as Agent Canvas, a self-hosted, always-on developer control center that turns coding agents into an engineering team. https://github.com/OpenHands/OpenHands (checked 2026-11-06)
  112. OpenHands is an MIT-licensed open-source, self-hosted control center for coding agents and automations. https://github.com/OpenHands/OpenHands (checked 2026-11-06)
  113. Agent Canvas automations run on a schedule or on webhook events, with examples such as publishing reports to Slack and decomposing GitHub issues into tasks. https://github.com/OpenHands/OpenHands (checked 2026-11-06)
  114. Pressing & or running /handoff in a local Warp Agent conversation forks the transcript and snapshots uncommitted changes, including untracked files, into a cloud run. https://docs.warp.dev/platform/handoff/local-to-cloud/ (checked 2026-11-06)
  115. Warp's oz-agent-action GitHub Action runs agents from workflow triggers such as @oz-agent comments, pull request events, issue labels, daily schedules or CI failures, using a Warp API key stored as a repository secret. https://docs.warp.dev/platform/integrations/github-actions (checked 2026-11-06)
  116. Other public ACP registry agents run in Nori only when the ACP registry feature is explicitly enabled, and Gemini was removed from managed Sessions and is rejected by the broker. https://docs.noriagentic.com/sessions/reference/supported-agents-and-compute (checked 2026-11-06)
  117. Amazon's 2026-09-29 press release describes Kiro as an agentic AI-powered software development environment from Amazon Web Services that offers spec-based coding, where agents write and verify specifications before generating any code. https://press.aboutamazon.com/aws/2026/9/tcbs-goes-all-in-with-kiro-amazon-web-services-agentic-coding-environment (checked 2026-11-06)
  118. Kiro Web launched on 2026-05-18 for paid subscribers with an Autonomous toggle under which Kiro asks clarifying questions, drafts a plan, uses sub-agents to plan, write and verify changes, and returns a pull request; it was the first product to ship the autonomous vision AWS presented at re:Invent. https://kiro.dev/blog/introducing-kiro-web/ (checked 2026-11-06)
  119. Kiro cloud sessions run in a managed, isolated cloud sandbox with the same agent harness as the IDE and CLI, up to 10 at a time, with no separate charge for cloud compute; kiro-cli --cloud starts one from a terminal and the IDE lists them in Agent Focus Mode. https://kiro.dev/docs/cloud-sessions.md (checked 2026-11-06)
  120. Kiro Web starts work when a GitHub issue gets the kiro label or a /kiro comment, pushes a feature branch, opens a pull request with a detailed description, cannot push to protected branches, and leaves review and merging to your normal GitHub workflow. https://kiro.dev/docs/web/github.md (checked 2026-11-06)
  121. Kiro's plans are Free (50 credits a month), Pro at $20 with 1,000 credits, Pro+ at $40 with 2,000, Pro Max at $100 with 5,000 and Power at $200 a month per user with 10,000 credits, with add-on credits at $0.04 each on paid tiers; Kiro Web needs a paid plan. https://kiro.dev/pricing/ (checked 2026-11-06)
  122. Kiro's FAQ says it works as an agent in the IDE, with a CLI, in a web browser, from a mobile app and in Kiro Crew, and that Kiro Web can connect GitHub or GitLab repositories or both. https://kiro.dev/faq/ (checked 2026-11-06)
  123. Kiro offers Claude, GPT-5.6 and open-weight models such as Kimi K3, GLM-5, DeepSeek 3.2 and Qwen3 Coder Next through its own credits, charged at multipliers relative to Auto at 1.0x, for example Claude Opus 5.5 at 2.0x, Claude Sonnet 5.5 at 1.3x and GPT-5.6 Sol at 4.4x. https://kiro.dev/docs/models.md (checked 2026-11-06)
  124. Google's Jules changelog entry dated 2026-10-29 says Jules will wind down on 2026-12-31, that users have until 2026-11-11 to finish sessions, and that after 2026-11-11 Jules turns read-only until the end of the year. https://jules.google/docs/changelog/ (checked 2026-11-06)
  125. Google's turndown notice points to the Jules Tools CLI, installed with npm install -g @google/jules and signed in with jules login, plus an agent prompt that exports account settings, environment variables, scheduled tasks, connected repositories, memories, setup scripts and every session with its prompt, plan, activity log and code patches; code, branches and pull requests stay in GitHub. https://jules.google/docs/changelog/ (checked 2026-11-06)
  126. Google's turndown notice says what people built with Jules is shaping products across Google, including its agent-first platform, Google Antigravity. https://jules.google/docs/changelog/ (checked 2026-11-06)
  127. Google Antigravity describes itself as an agentic development platform with an agent manager, and says it is available to developers at no charge. https://antigravity.google/ (checked 2026-11-06)
  128. Jules runs on Google's Gemini models. https://jules.google/docs/usage-limits (checked 2026-11-06)
  129. Paid Jules tiers are offered to personal Google accounts; Workspace users are directed to an interest form. https://jules.google/docs/usage-limits (checked 2026-11-06)
  130. Anthropic reports that across enterprise deployments Claude Code averages about $13 per developer per active day and $150 to $250 per developer per month, staying below $30 per active day for 90% of users. https://code.claude.com/docs/en/costs (checked 2026-11-06)
  131. The legacy Claude in Slack bot was retired on 2026-10-05 and no longer responds in channels of workspaces connected to a Claude organization. https://code.claude.com/docs/en/slack (checked 2026-11-06)
  132. The claude setup-token command generates a one-year token tied to a Pro, Max, Ultra, Team or Enterprise plan for machines where browser login isn't available. https://code.claude.com/docs/en/authentication (checked 2026-11-06)
  133. Cognition introduced Devin Desktop as the next generation of Windsurf on 2026-06-02. https://cognition.com/blog/introducing-devin-desktop (checked 2026-11-06)
  134. Amp removed ads from its free tier on 2026-03-30. https://ampcode.com/news/amp-free-is-ad-free (checked 2026-11-06)
  135. Nori is led by CEO Amol Kapoor, a former Google Research scientist, and CTO and co-founder Clifford Ressel. https://ai.engineer/speakers/amol-kapoor (checked 2026-11-06)
  136. Amol Kapoor's Google-affiliated research includes MixHop (ICML 2019), N-GCN and a 2020 spatio-temporal graph neural network for COVID-19 forecasting. https://scholar.google.com/citations?user=SC_-qB8AAAAJ (checked 2026-11-06)
  137. Clifford Ressel is Nori's CTO and co-founder. https://ai.engineer/orgs/nori (checked 2026-11-06)
  138. Corvus's CTO saw a production bug at a client dinner, had Nori look into it and deploy a fix. https://noriagentic.com/newsletter/2026-05-19-corvus-case-study.html (checked 2026-11-06)