# Hand a whole task to a hosted agent and collect a pull request

> Hand a whole task to a hosted agent and collect a pull request: 1 companies fully do it, 10 with a limit, 48 carry the tag. Every listing in Agent harnesses with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/agent-harnesses.cloud-delegate/
- Markdown: https://www.anchorterminal.com/niches/agent-harnesses.cloud-delegate/index.md (~2,600 tokens)
- Slim: https://www.anchorterminal.com/niches/agent-harnesses.cloud-delegate/index.min.md (~380 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/agent-harnesses.cloud-delegate/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

One company's docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 1 | 10 | 48 | 3 | 2 |

## What counts as done

A branch or pull request appears, the plan was shown before execution when asked, and the task count or credit spend is visible afterwards.

Measured, for our own tests: the branch or pull request's checks pass.

How agents ask for it: open a PR that bumps these dependencies; fix this issue while I'm away; queue these five tasks

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Warp (Warp Agent CLI and the Oz Agent API) (Denver Technologies, Inc. (d/b/a Warp))**: Factories route an issue through triage, planning, implementation and review with human decision points and return a pull request to the thread; runs expose `state` and usage is billed to credits.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Amp (Amp Frontier Corporation)**: amp -ox runs the prompt in an orb on Amp's servers and returns the thread URL at once, but no plan approval step or pull request output is documented on the pages read.
- **Claude Code (Anthropic, PBC)**: Cloud routines run without the machine on and start in auto mode, and a label-driven pipeline merged a one-line change, but the cloud sandbox had no database so a UI change shipped unrendered (acc-07).
- **Devin (Cognition AI, Inc.)**: Sessions clone the repository, work in a VM and open pull requests with a max_acu_limit, but no idempotency key exists on creation and the 24 September 2026 outage took the web app down for two hours.
- **GitHub Copilot CLI (GitHub, Inc.)**: Cloud sandboxing runs a whole session in a GitHub-hosted ephemeral Linux environment billed by usage, and --connect joins a remote session, but the cloud agent's pull request flow was not read here.
- **HarnessRouter (Lumentree Corporation)**: A hosted agent runs the task and spend shows per task, but no pull request output or plan approval step is documented.
- **Kiro CLI (Amazon Web Services, Inc.)**: `--cloud --repo owner/repository` starts a managed cloud session from the headless page, but the cloud sessions page, plan approval and credit visibility were not read.
- **Mistral Vibe CLI (Mistral AI (RCS Paris 952 418 325))**: An `&` prefix sends a prompt to a Vibe Code Web cloud session that works on a GitHub repository, but the Web pages were not read and this listing covers the CLI.
- **OpenAI Agents API (OpenAI OpCo, LLC)**: OpenAI hosts the agent and sandbox, but no GitHub clone or pull request flow is documented, so an agent would push with its own token; billing is tokens plus container minutes, not task credits.
- **OpenAI Codex (OpenAI OpCo, LLC)**: `codex cloud exec` submits a task and `codex cloud list --json` returns `status` per task, but cloud needs a ChatGPT plan rather than an API key and acc-31 found its automatic PR creation less reliable than Jules.
- **OpenHands (All Hands AI)**: OpenHands Cloud automations clone with your git login and can open issues or PRs, but the Cloud task pages, plan approval and credit display were not read.
- **Parallax (Conifer Solutions, Inc.)**: The hosted service opens one fix PR per review with proven patches on GitHub and spends episode credits (5 free, then one per episode), but it reviews work rather than taking a task, and the hosted door has no spend cap.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **Cursor CLI (Anysphere, Inc.)**: The listing's bestFor names a hand-off to Cloud Agents, but the cloud agents docs were not opened in this read.
- **Qoder CLI (BRIGHT ZENITH PRIVATE LIMITED)**: docs.qoder.com answered 502 from the fetch proxy on three attempts and a DNS failure through WebFetch, so no page was read.
- **Unstoppable (Unstoppable Domains, Inc.)**: Cloud tasks and MCP tools to wait on them appear in the 1.15.0 changelog with no documentation or pricing.

- not this job: Agent Zero (Agent Zero, s.r.o.): Self-hosted in Docker; no vendor runner.
- not this job: Aider (Aider AI LLC): Runs locally; no hosted runner.
- not this job: Antigravity CLI (Google LLC): agy runs on the user's machine; no hosted runner that clones a repository and opens a pull request is documented.
- not this job: Auggie CLI (Augment Computing, Inc.): Auggie runs in your CI or terminal; the GitHub Actions wrapper runs on your runners, not a vendor-hosted agent that returns a pull request.
- not this job: AXLR (Underpass AI): No hosted AXLR exists; the service runs on a machine the operator controls.
- not this job: Clawboo (Sanreds): Self-hosted only; no hosted option is offered.
- not this job: CLI Manager (solhun (Jeong Kyeonghun)): Local macOS app only.
- not this job: Cline (Cline Bot Inc.): The CLI, extension and hub run on your machines; no vendor-hosted runner returns a pull request.
- not this job: Conductor (Conductor): Conductor Cloud goes read-only on 13 November 2026 and shuts down on 20 November 2026.
- not this job: Crush (Charmbracelet, Inc.): Crush runs locally; no hosted runner.
- not this job: DeepSeek Harness (dsh) (DeepSeek (the LICENSE names DeepSeek, and no legal entity is named in the repository or the docs)): Runs locally; no hosted runner.
- not this job: dsh-superagent (DeepSeek) (DeepSeek (the LICENSE names the DeepSeek Harness Team; DeepSeek Harness's terms name Hangzhou DeepSeek Artificial Intelligence Co., Ltd.)): Self-hosted on the operator's machine; no vendor runner.
- not this job: Droid CLI (The San Francisco AI Factory, Inc.): The CLI runs on your machine; no hosted runner that returns a pull request was found in the pages read.
- not this job: Pi (Earendil Inc.): Runs locally; Radius is a model gateway, not a runner.
- not this job: Emdash (General Action, Inc.): Local and SSH machines only; no vendor runner.
- not this job: Gemini CLI (Google LLC): Runs locally; no hosted runner.
- not this job: goose (Agentic AI Foundation (Linux Foundation)): Runs locally or on your own server; no vendor runner.
- not this job: Harness (Autonomous, Inc.): Runs on the person's own computers.
- not this job: Harness² (Google LLC): No hosted service.
- not this job: Jules (Google LLC): Jules showed a plan, published a branch and left the PR to the author with five tasks a day (acc-31), but sessions are read-only after 11 November 2026 and the service ends on 31 December.
- not this job: Junie CLI (JetBrains s.r.o.): Runs locally; no hosted runner.
- not this job: Kilo Code CLI (Kilo Code Inc. (the terms are a contract with Anaconda, Inc. on behalf of itself and its affiliates, including Kilo Code Inc.)): Runs locally; the Kilo Gateway routes models, not tasks.
- not this job: Kimi Code CLI (Moonshot AI (the global Kimi terms name NOVASCENT PRIVATE LIMITED, Singapore)): No hosted runner; the CLI runs on the user's machine or a declared remote environment.
- not this job: Letta (Letta): `--computer cloud` runs a prompt in a Letta sandbox, but no repository or pull request flow is documented.
- not this job: mini-swe-agent (No legal entity named. The MIT licence names Kilian A. Lieret and Carlos E. Jimenez): No hosted runner; Modal and Docker environments are for batch evaluation, not a pull request service.
- not this job: MiniMax Code (Nanonoble Pte. Ltd.): No hosted runner for the CLI.
- not this job: Muse Code (Meta Platforms, Inc.): No hosted runner.
- not this job: OpenCode (Anomaly Innovations, Inc.): No hosted runner; Zen and Go are model subscriptions.
- not this job: OpenMausBot (Supamaus Software Private Limited): MausBot Cloud hosts a workspace, but no pull request flow is documented.
- not this job: Orca (Lovecast Inc.): Cloud VMs host worktrees you control; no pull request service is described.
- not this job: Paperclip (Paperclip Labs, Inc.): Paperclip Cloud is a waitlist with no published service.
- not this job: Paseo (Mohamed Boudra): Agents run on your own machines; no hosted service.
- not this job: Prime Agent (Prime Intellect, Inc.): No hosted runner.
- not this job: Qwen Code (Qwen team, Alibaba Group (no legal entity is named in the repository or the docs)): No hosted runner.
- not this job: Mirror CLI (Reflection AI): No hosted runner.
- not this job: Rovo Dev CLI (Atlassian Pty Ltd): No hosted task runner; Bitbucket Pipelines integration runs in your CI.
- not this job: Sculptor (Imbue, Inc.): The container backend is experimental and self-managed; no hosted PR service.
- not this job: T3 Code (T3 Tools Inc.): Agents run on your machines.
- not this job: Traycer (Traycer AI): Agents run on your hosts; Traycer inference is only the model.
- not this job: Verdent (Verdent AI, Inc.): Verdent Cloud builds web apps in the browser; no repository task delegate is documented.
- not this job: Vibe-Trading (HKUDS): Self-hosted only.

## What stops agents finishing

- Jules turns read-only after 11 November 2026
- no idempotency on session creation so a retry doubles the work
- daily task caps on free tiers (acc-31)
- no database in the sandbox so UI changes cannot be rendered (acc-07)

## Around it

Before: issue-tracking.read-issue. After: code-hosting.review-pr.

## Where this comes from

Weight 3 of 100 in Agent harnesses: 2 of 33 accounts (acc-07, acc-31); Devin and Codex cloud produced only vendor numbers

As data: https://www.anchorterminal.com/api/v1/use-cases/agent-harnesses.cloud-delegate.json, or the MCP tool get_use_case with id agent-harnesses.cloud-delegate. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
