MonkeyCode

Self-hosted coding platform, unlike Cursor cloud requests

by Chaitin Tech

Overview

MonkeyCode is Chaitin open-source AI development platform, with hosted tasks, private deployment and a separate Work desktop client. The platform code is AGPL-3.0, and the official documentation covers binding Git platforms, repositories and branches, running tasks in a development environment and pushing results to a remote or downloading them locally. Model endpoints, infrastructure and plan entitlements decide where data goes and what it costs: a private deployment guide asks for a Linux control node of at least 2 cores, 4GB and 40GB plus a development host recommended at 8 cores, 16GB and 100GB, and requires you to bind a model API address, token and model name, while the current guide caps member accounts at five. Keeping files on your own machines does not make the model calls local, so a fully offline setup also needs models, Git, packages and tools available inside the permitted environment. On price, the online guide lists a free basic tier, Professional at CNY 99 a month or CNY 999 a year and Flagship at CNY 499 a month or CNY 4,999 a year, in yuan rather than dollars, with Professional granting 10,000 credits a month and Flagship 100,000. The free allowance is where the vendor disagrees with itself: one guide states 10 million tokens a day and a separate allowance page states 30 million basic-model tokens, so we report the conflict rather than pick one. The daily basic-model token allowance refreshes without accumulating, and exhausting it can deduct points from the separate points balance, which tool calls and some models also draw on. Work separates local tasks, cloud tasks and standalone chat; local tasks read and write an authorised folder but still need a model configured, which can be the cloud account, BaiZhi Cloud or a custom endpoint, and BaiZhi Cloud credits are a separate balance from MonkeyCode credits.

MonkeyCode answers the training question directly in its own FAQ: your repositories, prompts and outputs do not enter any model training pipeline by default. Its user agreement carries no training clause, and its privacy policy gives retention only as the shortest period needed for the business purpose, without a number (MonkeyCode FAQ, user agreement and privacy policy, read 12 October 2026).

Editor's take

A fit for teams that want to run an AI development platform inside their own infrastructure and keep project code in Git. Check which model endpoint the work will use, since that decides both the data path and the cost; the hosted service and the private deployment have different entitlements, and the free allowance is documented inconsistently.

Our reading of who this suits, based on the sources above. Opinion, offered to help you decide - not a lab result.

Direct comparisons

Cursor vs MonkeyCode

Key Features

•Open-source platform under AGPL-3.0, with hosted tasks and private deployment
•Bind Git platforms, repositories and branches, run tasks in a development environment, push to a remote or download files locally
•Private deployment guide: control node at least 2 cores, 4GB and 40GB; development host recommended at 8 cores, 16GB and 100GB; member accounts capped at five in the current guide
•Free basic tier; Professional CNY 99 a month or CNY 999 a year with 10,000 credits a month; Flagship CNY 499 a month or CNY 4,999 a year with 100,000
•The daily basic-model token allowance refreshes without accumulating; exhausting it can deduct points from the separate points balance
•Model endpoint, token and model name are configured by the operator, so data routing and cost follow that choice

Use Cases

→Running AI coding tasks on your own infrastructure
→Task and team management around Git repositories
→Local desktop tasks on authorised folders
→Publishing a project through the platform web app

Pros

✓The platform code is open source under AGPL-3.0 and can be deployed on your own infrastructure
✓Git workflows and local file download help keep project source under your control
✓A separate Work desktop client covers local tasks, cloud tasks and standalone chat

Cons

✕The free daily allowance is documented inconsistently: 10 million tokens in one guide against 30 million basic-model tokens in another
✕A deployment needs a configured model endpoint: a remote endpoint sends context off-host, while local inference needs a local model setup
✕The current deployment guide caps member accounts at five
✕Long-term hosting commitments for published web apps - production SLA, custom domain and backup - were not verified in this round
✕BaiZhi Cloud credits and MonkeyCode credits are separate balances

Frequently Asked Questions

Is MonkeyCode free?

There is a free basic tier, but the hosted service is not free in general: the online guide lists Professional at CNY 99 a month or CNY 999 a year and Flagship at CNY 499 a month or CNY 4,999 a year, and top models and tools are not covered by the basic allowance. The free daily allowance is also documented inconsistently - 10 million tokens in one guide and 30 million basic-model tokens in the allowance page - so use what your account currently shows.

Does private deployment mean nothing leaves my network?

Not by itself. The private deployment guide requires binding a model API address, token and model name, so inference follows that endpoint. Control plane and files stay local, but a fully offline setup also needs models, Git, packages and tools available inside the permitted environment; we did not test a disconnected install.

Is MonkeyCode a completion plugin in my IDE?

The current official repository describes an AI development platform with cloud environments, tasks and team management plus private deployment, so the completion-assistant framing does not match the current product. A separate Work desktop client covers local tasks, cloud tasks and standalone chat.

What does publishing a web app include?

The online guide says the platform builds and packages the app and generates a link, with publishing waiting for administrator review and later updates going through a ticket. That is not a production SLA, a custom domain or a database backup guarantee, and we did not verify long-term hosting constraints.

What limits apply to the daily allowance and invitations?

The daily basic-model token allowance refreshes without accumulating, and exhausting it can deduct points from the separate points balance rather than resetting a large daily credit pool. Invitations are capped by the announcement of 2 September 2026 at ten rewards per person and 50,000 points in total, which supersedes the older unlimited-invite wording still present on the credits page.

How does MonkeyCode compare to Cursor?

MonkeyCode can be deployed on your own infrastructure and prices Professional at CNY 99 a month; Cursor’s US$20 Pro price applies to eligible legacy plans; new customers use The X Subscription and its applicable price remains unestablished here. Cursor’s documented standard workflow sends requests through its backend. Private deployment does not by itself keep inference local, because the model endpoint still decides that path. This is a difference in where the platform runs, not a measured difference in code quality.

Key facts and where they come from

PriceNot established: The registered record gives plan amounts and billing cadence, but the applicable purchase market/region and full account/channel conditions were not established. These are not asserted as a current global purchasable quote.Source: MonkeyCode - online guide
AvailabilityNot established: Entry, integration or deployment routes are registered, but the applicable geographic access conditions were not established. No worldwide availability is inferred from these routes.Source: chaitin/MonkeyCode repository · MonkeyCode - private deployment guide
Commercial useDocumented: The platform code is AGPL-3.0. The licence covers the platform itself, not the dependencies, model services or commercial hosting of generated projects.Source: MonkeyCode licence (AGPL-3.0)
Training on your dataDocumented: Not by default: your repositories, prompts and outputs do not enter any model training pipeline by defaultSource: MonkeyCode - FAQ on model training
API billingNot established: The Work source separates provider-billed model endpoints, SaaS credits and BaiZhi Cloud balances; MonkeyCode’s own programmatic API billing and consumer-plan coverage remain unestablished.Source: MonkeyCode Work
Export formatsNot established: Pushing task results to a remote repository or downloading them locally is recorded, but archive/file format and export restrictions were not established.Source: MonkeyCode - online guide
Usage limitsDocumented: States a free allowance of 30 million basic-model tokens a day, plus 100 million for Professional and 300 million for Flagship, where the online guide states 10 million for the free tier - a conflict we report rather than resolve. The daily basic-model token allowance refreshes without accumulating, and exhausting it can deduct points from the separate points balance. From 2 September 2026 each person is limited to ten invitation rewards and 50,000 credits in total, which replaces the unlimited-invite wording. Campaign rewards should not be used as a plan budget.Source: MonkeyCode - credit rules · MonkeyCode - invitation cap (2 September 2026)

Official sources

The vendor pages we read for the claims on this page, what each one states, and the day we read it.

chaitin/MonkeyCode repository ↗
The official main repository presents an AI development platform with cloud environments and task and team management, supporting self-deployment; it does not describe an IDE completion assistant, which is why this entry no longer uses that framing.
Read on September 23, 2026
MonkeyCode licence (AGPL-3.0) ↗
The platform code is AGPL-3.0. The licence covers the platform itself, not the dependencies, model services or commercial hosting of generated projects.
Read on September 23, 2026
MonkeyCode - online guide ↗
Binding Git platforms, repositories and branches, running tasks in a development environment, and pushing results to a remote repository or downloading them locally. Plans: basic free, Professional CNY 99 a month or CNY 999 a year, Flagship CNY 499 a month or CNY 4,999 a year, with 10,000 and 100,000 credits a month and 1 and 3 concurrent tasks. Web app publishing builds and packages the app and generates a link, subject to administrator review. It also states a free allowance of 10 million tokens a day, which the separate allowance page contradicts.
Read on September 23, 2026
MonkeyCode - private deployment guide ↗
Requires a Linux control node of at least 2 cores, 4GB and 40GB and a development host recommended (not required) at 8 cores, 16GB and 100GB, and binding a model API address, token and model name; the current guide caps member accounts at five and points larger commercial needs to the vendor. Keeping the control plane local does not make external API calls local.
Read on September 23, 2026
MonkeyCode - credit rules ↗
States a free allowance of 30 million basic-model tokens a day, plus 100 million for Professional and 300 million for Flagship, where the online guide states 10 million for the free tier - a conflict we report rather than resolve. The daily basic-model token allowance refreshes without accumulating, and exhausting it can deduct points from the separate points balance.
Read on September 23, 2026
MonkeyCode - credit boundaries ↗
Basic cloud resources and basic models are free, but arbitrary top models and tools are not; the page still carries an unlimited-invite line that the newer announcement supersedes.
Read on September 23, 2026
MonkeyCode - invitation cap (2 September 2026) ↗
From 2 September 2026 each person is limited to ten invitation rewards and 50,000 credits in total, which replaces the unlimited-invite wording. Campaign rewards should not be used as a plan budget.
Read on September 23, 2026
MonkeyCode Work ↗
Work separates local tasks, cloud tasks and standalone chat. Local tasks read and write an authorised folder but still need a model configured, which may be the cloud account, BaiZhi Cloud or a custom endpoint, billed by that service; cloud tasks use the SaaS membership and credits. BaiZhi Cloud credits and MonkeyCode credits are two separate balances.
Read on September 23, 2026
MonkeyCode - FAQ on model training ↗
Read on 12 October 2026. The FAQ answers the training question directly, verbatim: "代码会不会被拿去训练模型？默认不会。你的仓库、prompt 和输出默认不进入任何模型训练流程。自托管版本数据也不会出你的网络。" - will the code be used to train models? Not by default: your repositories, prompts and outputs do not enter any model training pipeline by default, and self-hosted data does not leave your network either.
Read on October 12, 2026
MonkeyCode - privacy policy ↗
Read on 12 October 2026. Retention is given as a standard, not a number, verbatim: "您的信息会在实现业务目的所需的最短期限内保存，超期后将按规则删除或匿名化处理。" - your information is kept for the shortest period needed for the business purpose, and deleted or anonymised under the rules once that period expires. The policy contains no clause about using user content to train models. The /privacy-policy route is client-rendered and its text was read from the site's own bundle (assets/index-DgOWGWHy.js).
Read on October 12, 2026
MonkeyCode - user agreement ↗
Read on 12 October 2026. The agreement has no model-training clause and states no retention period. On user content it says only that, to provide the service, you authorise the platform to store, process, transmit and display the content as far as necessary, verbatim: "为向您提供服务，您授权平台在必要范围内对相关内容进行存储、处理、传输与展示。" The /user-agreement route is client-rendered and its text was read from the site's own bundle.
Read on October 12, 2026

Compare with other tools in Code Tools

Product	Maker	Pricing	Launched	Shared tags
Claude Code	Anthropic	Paid	2025-02-24	Coding
Devin Desktop	Cognition	Freemium	2024-11-13	Coding
Replit Agent	Replit	Freemium	2024-09	Coding
Cursor	Cursor	Freemium	2023-03	Coding
GitHub Copilot	GitHub	Freemium	2021-10	Coding
Tabnine	Tabnine	Paid	2018-06	Coding