Guide · Agent skill

How to use the Atomic Chat Local Models agent skill

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

This skill connects a NanoClaw agent to models you run yourself in the Atomic Chat desktop app. Claude stays in charge of the task, but it can list the models you have installed and send them prompts, getting answers back from your own machine. This guide explains what an agent skill is, how to install this one, how to trigger it, and what it changes in your code.

Atomic Chat Local Modelsnanocoai/nanoclaw · MIT Download the skill

Quick start

1Check Atomic Chat. Install Atomic Chat on macOS, enable the Local API Server on port 1337, and download at least one model.
2Unzip the skill. Download the skill ZIP and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
3Apply the code changes. Run the skill in your NanoClaw repository so it copies the files and edits index.ts, container-runner.ts and the Docker driver.
4Run the build and tests. Run pnpm run build, the TypeScript check, the two test files and ./container/build.sh, and make sure all pass.
5Configure the host. Optionally set ATOMIC_CHAT_HOST in .env, and leave ATOMIC_CHAT_API_KEY unset for a default local install.
6Restart and ask. Restart the NanoClaw service, then ask the agent to list or use your local models.

What is an agent skill, and what does this one do?

An agent skill is a folder of instructions and files that an AI agent can read and follow to gain a new capability. This skill, add-atomic-chat-tool, is designed for the NanoClaw agent and adds a small stdio-based MCP server that exposes local models as tools.

It gives Claude two tools. atomic_chat_list_models lists the models currently available in Atomic Chat, and atomic_chat_generate sends a prompt to a chosen model and returns the response. Requests go to the Atomic Chat local OpenAI-compatible API at http://127.0.0.1:1337/v1. Model downloads and deletions are still done in the Atomic Chat desktop UI.

atomic_chat_list_models: calls GET /v1/models
atomic_chat_generate: calls POST /v1/chat/completions

Install the skill: download the ZIP and unzip it

Download the skill ZIP from the directory page, then unzip it into one of two locations. Use ~/.claude/skills/ for a personal install, or a project's .claude/skills/ folder to scope it to one project.

Alternatively, if you use the Claude apps, you can upload the skill under Settings → Capabilities → Skills.

Personal: unzip into ~/.claude/skills/
Per project: unzip into your project's .claude/skills/
Claude apps: Settings → Capabilities → Skills, then upload

Check the prerequisites before applying the skill

The skill needs the Atomic Chat desktop app (macOS only, per the source instructions) with its Local API Server enabled on port 1337. Open Settings → Local API Server to confirm it is on, then download at least one model from the Hub or Models tab.

Applying the skill also requires a NanoClaw repository with the container/agent-runner and Docker setup, plus Bun and Node/pnpm for builds and tests. Load a model once by sending a message in Atomic Chat so it warms up.

Atomic Chat installed and running on macOS
Local API Server enabled on port 1337
At least one downloaded model
NanoClaw repository with Docker setup
Bun and Node/pnpm toolchain

curl -s http://127.0.0.1:1337/v1/models | head

Apply the code changes NanoClaw needs

Applying the skill copies four files into the NanoClaw tree: the MCP server, the env-forwarding helper, and two tests. It then edits index.ts to register the atomic_chat server, edits container-runner.ts to forward env vars, and changes the [ATOMIC] log branch in the Docker driver's stderr handler. Registering the server is what makes its tools visible to the agent.

The source notes that this edits several integration points and a shared logger block that other local-model skills also touch. Run the build and tests listed in the skill before continuing.

Copies atomic-chat-mcp-stdio.ts and its registration test into container/agent-runner/src/
Copies atomic-chat-env.ts and its wiring test into src/
Adds an atomic_chat entry to the mcpServers object in index.ts
Adds ...atomicChatEnv() to contributedEnv in composeSessionSpec
Changes the stderr handler in src/drivers/docker-driver.ts so [ATOMIC] lines log at info

pnpm run build
pnpm exec tsc -p container/agent-runner/tsconfig.json --noEmit
pnpm exec vitest run src/atomic-chat-wiring.test.ts
(cd container/agent-runner && bun test src/atomic-chat-registration.test.ts)
./container/build.sh

Configure the host and restart the service

By default the server connects to http://host.docker.internal:1337 with a fallback to localhost. To use a different host, set ATOMIC_CHAT_HOST in .env. Atomic Chat does not require authentication when run locally, so leave ATOMIC_CHAT_API_KEY unset unless you have placed it behind a reverse proxy that enforces auth.

After changing configuration, restart the NanoClaw service from your project root. The source gives the macOS command below; on Linux it uses systemctl --user restart.

ATOMIC_CHAT_HOST: optional custom host
ATOMIC_CHAT_API_KEY: optional, only behind an auth proxy

source setup/lib/install-slug.sh
launchctl kickstart -k gui/$(id -u)/$(launchd_label)

Trigger the skill with natural requests

You do not need special syntax. Ask the agent in plain language to use your local models for a task, and name the model if you want a specific one. The agent decides when to call the tools, so the requests below are examples of the kind of wording that fits.

Because Claude remains the orchestrator, it still plans the work and reads the answers. Local offloading suits simple, self-contained prompts.

"List the models I have available in Atomic Chat."
"Use my local Qwen model to summarise this paragraph."
"Send this classification question to a local model and tell me what it says."

Safety notes: scripts and network access

This skill runs scripts and uses the network. It modifies core NanoClaw source files and Docker-related logging code, so review the diff before building. Requests go to the local Atomic Chat API, not a cloud service, but the host and key settings control where traffic is sent.

To undo the changes, follow the REMOVE.md file in the skill folder. It deletes the copied files, removes the atomic_chat entry, reverts the host-side edits and removes the env var block. The removal steps are safe to run even if some steps were never applied.

Review the file edits before building
Only set ATOMIC_CHAT_API_KEY if you use an auth proxy
Use REMOVE.md to revert cleanly

What people use it for

Listing which models are installed in Atomic Chat before choosing one for a task
Offloading simple summarisation or rewriting prompts to a local model
Keeping sensitive text on your own machine by sending it to a local model instead of a cloud service
Testing how a local model answers a prompt you plan to use in a NanoClaw workflow

Frequently asked questions

Is the Atomic Chat local models skill free?

The skill is published under the MIT licence, as stated on its page. Atomic Chat is a separate desktop app, and you should check its own release page for its terms.

Does this skill work on Windows or Linux?

The source says Atomic Chat is macOS only for now, so the skill's local setup is documented for macOS. The restart command for Linux in the source uses systemctl --user restart.

Do I need an API key for Atomic Chat?

No. The source says Atomic Chat does not require authentication when running locally. Only set ATOMIC_CHAT_API_KEY if you have placed it behind a reverse proxy that enforces auth.

Why can't the agent see the Atomic Chat tools?

The tools only appear if the atomic_chat server is registered in index.ts. The registration test checks this entry exists. Run the tests and confirm the index.ts edit is in place.

How do I download models for Atomic Chat?

Model management is done in the Atomic Chat desktop UI, using the Hub or Models tab. The skill does not download or delete models itself.

How do I remove the skill?

Follow the REMOVE.md file in the skill folder. It deletes the copied files, unregisters the atomic_chat server, reverts the host-side edits and removes the env var block from .env.example.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

This skill suits developers running NanoClaw on macOS who already use Atomic Chat and want to offload simple prompts to a local model. It is not a good fit if you do not have Atomic Chat running on port 1337, or if you are unwilling to let a skill edit several core NanoClaw source files and a shared logger block.

Download the skill