What is ultraworkers/claw-code?

Claw Code is the public Rust implementation of the claw CLI agent harness, built and maintained through AI agents rather than direct human commits. Its own README describes it as a museum exhibit rather than a production product, and explicitly points new users toward LazyCodex and Gajae-Code for real work.

Can I use ultraworkers/claw-code commercially?

Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.

Is ultraworkers/claw-code still maintained?

Yes. The repository last received commits 44 days ago.

What programming language is ultraworkers/claw-code written in?

Mainly Rust, according to GitHub's language statistics.

Claw Code: Open-Source Rust CLI Agent Harness

At a glance

What is it?: Claw Code is the public Rust implementation of the claw CLI agent harness, built and maintained through AI agents rather than direct human commits. Its own README describes it as a museum exhibit rather than a production product, and explicitly points new users toward LazyCodex and Gajae-Code for real work.
Who is it for?: Developers curious about how a Rust CLI agent harness operates, or who want to study a codebase managed by AI agents, will find Claw Code an unusual but documented example. Engineers who need a working agent harness for daily coding tasks should go to LazyCodex or Gajae-Code as the README recommends.
Can I use it commercially?: Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?: Yes. The repository last received commits 44 days ago.
What is it written in?: Mainly Rust, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on September 29, 2026, and from our analysis. They are not legal advice.

DEEP OPEN-SOURCE ANALYSIS

Hysen Labs Editorial · August 17, 2026 · ultraworkers/claw-code

What Claw Code Is and Why the README Calls It a Museum Exhibit

Claw Code is the public Rust implementation of the `claw` CLI agent harness. The repository is part of the ultraworkers project, which uses the Gajae-Code and LazyCodex agent frameworks to plan, execute, verify, label, and preserve the codebase without direct human intervention in the usual sense. The agents run the harnesses; the harnesses run the agents.

The README is direct about what this is not: a product pitch. It describes the repository as closer to a museum exhibit than a deployment target, an artifact kept alive by automated agents rather than a conventional maintainer. The README directs anyone who wants to run real workloads to LazyCodex at github.com/code-yeongyu/lazycodex or Gajae-Code at github.com/Yeachan-Heo/gajae-code. Claw Code is for inspecting the artifact, not for daily use.

The last push was on 2026-08-16. The repository has no GitHub releases; versioning is not a published concept here. The MIT license applies to all code in the repository.

Repository Layout and the Key Documentation Files

The canonical Rust workspace lives in `rust/`. That is where the `claw` CLI binary is built, and the README calls it the current source of truth. Everything else is supplementary.

The top-level documentation files cover distinct concerns. USAGE.md is the task-oriented guide for build, authentication, CLI usage, session management, and parity-harness workflows. PARITY.md tracks the Rust-port checkpoint relative to earlier implementations. ROADMAP.md lists the active roadmap and cleanup backlog. PHILOSOPHY.md explains the project intent and system design.

The `src/` and `tests/` directories hold a companion Python and reference workspace. The README is explicit that these are not the primary runtime surface. The `docs/` directory contains specific guides: `docs/local-openai-compatible-providers.md` for local model setup, `docs/navigation-file-context.md` for file submission questions, `docs/windows-install-release.md` for Windows users, and `docs/container.md` for the container-first workflow.

Building from Source: The Only Reliable Installation Path

The README includes a prominent warning about crates.io: `cargo install claw-code` installs a deprecated stub that places `claw-code-deprecated.exe`, not `claw`. Running that stub only prints `"claw-code has been renamed to agent-code"`. It is not the tool.

For the upstream binary without building from source, the README gives:

bash

cargo install agent-code

This installs `agent.exe` on Windows or `agent` on Unix. For the full build from the ultraworkers repository:

bash

git clone https://github.com/ultraworkers/claw-code
cd claw-code/rust
cargo build --workspace

After the build, set the Anthropic API key (this is the API key from platform.anthropic.com, not a Claude subscription):

bash

export ANTHROPIC_API_KEY="sk-ant-..."

Then verify the wiring:

bash

./target/debug/claw doctor

The README recommends making `claw doctor` the first health check after building. Once it passes, run a test prompt:

bash

./target/debug/claw prompt "say hello"

And to open an interactive session:

bash

./target/debug/claw

Windows Setup via PowerShell

PowerShell is a documented Windows path. The README lists four common onboarding steps for Windows users. First, install Rust from rustup.rs and reopen the terminal after the installer finishes. Second, verify Rust is on PATH:

powershell

cargo --version

If that fails, reopen the terminal or run the PATH setup from the Rust installer output. Third, clone and build:

powershell

git clone https://github.com/ultraworkers/claw-code
cd claw-code/rust
cargo build --workspace

Fourth, set the API key and run a test:

powershell

$env:ANTHROPIC_API_KEY = "sk-ant-..."
.\target\debug\claw.exe prompt "say hello"

The README notes that on Windows the binary is `claw.exe`, not `claw`, and that paths use backslashes in PowerShell. Alternatively, `cargo run -- prompt "say hello"` skips the path lookup entirely. For release ZIPs and PATH setup details, the README refers to `docs/windows-install-release.md`.

The RAG Service and Local Provider Support

The docker-compose.yml in the repository defines a three-service stack: a Qdrant vector database on ports 6333 and 6334, a RAG serve service that runs on port 8787, and a RAG ingest service. Both the serve and ingest services are built from `rust/crates/claw-rag-service/`. The CLAW_RAG_MOCK_PROVIDERS environment variable is set to `"1"` by default in the compose file, which means mock embeddings are used in local development and a real provider must be configured by overriding that variable in the shell.

The `CLAW_RAG_DB` variable points to a SQLite database at `/data/index.sqlite`. The ingest service mounts the repository root as a read-only workspace volume at `/workspaces/main`. This setup allows the agent to index the repository content and retrieve it during prompts.

For local LLM use, the README references `docs/local-openai-compatible-providers.md` for setup and states that offline skill installs are supported. The documentation covers how to configure a local OpenAI-compatible endpoint in place of the default Anthropic provider.

What Has Not Been Implemented and Known Naming Confusion

The ACP and Zed integration is incomplete. The README states that `claw-code` does not ship an ACP or Zed daemon or JSON-RPC entrypoint. Running `claw acp serve` currently returns a status message with exit code 0 and is a discoverability alias only. The ROADMAP.md tracks real ACP support separately, and the public JSON contract is documented in `docs/g011-acp-json-rpc-status-contract.md`. There is no timeline for completion stated in the README.

The naming situation on crates.io is a persistent source of confusion: the `claw-code` crate installs the wrong binary. This is described in the README's Quick start section, with a direct warning. The intended upstream binary is `agent-code`, which installs a binary named `agent` or `agent.exe`, not `claw`. The ultraworkers/claw-code repository must be built from source to get the `claw` binary.

LazyCodex, one of the harnesses used to maintain Claw Code itself, is the recommended starting point for engineers who want a working agent harness. LazyCodex and Gajae-Code are described in the README as the real tools for running work; Claw Code is the exhibit built on top of them.

Editorial conclusion

Developers curious about how a Rust CLI agent harness operates, or who want to study a codebase managed by AI agents, will find Claw Code an unusual but documented example. Engineers who need a working agent harness for daily coding tasks should go to LazyCodex or Gajae-Code as the README recommends. Before building, verify Rust is on PATH with `cargo --version` and set a valid ANTHROPIC_API_KEY; then run `./target/debug/claw doctor` to confirm the wiring is correct.

Frequently asked questions

What is the Claw code?

Claw Code is the public Rust implementation of the claw CLI agent harness, part of the ultraworkers project. Its own README describes it as an agent-managed exhibit maintained by AI agents rather than a production product, and directs users who need a working harness to LazyCodex or Gajae-Code.

How do you install Claw Code?

Build from the repository source by cloning ultraworkers/claw-code, navigating to the `rust/` directory, and running `cargo build --workspace`. The crates.io package `claw-code` is a deprecated stub; use `cargo install agent-code` for the upstream binary instead.

How do you use Claw Code locally with a local model provider?

The README references `docs/local-openai-compatible-providers.md` for configuring a local OpenAI-compatible endpoint and for offline skill installs. Set the local provider details in the configuration instead of using the default Anthropic API key path.

Which is better, Claude Code or Claw code?

Claude Code is Anthropic's official CLI agent. Claw Code is an independent Rust implementation described by its own README as a museum exhibit rather than a production tool, with the repository itself recommending LazyCodex for actual work. They serve different purposes.

Official sources

Community notes