Sandbox
@arcee-ai/nac

Long-running agent task harness for threads and episodes

nac gives agents a central orchestrator that plans work but does not directly edit files or run commands. Threads do the work and return structured episodes, so longer sessions stay aligned and easier to review.

234 stars20 forksRustUpdated 7d ago
Who it's for

Builders who want long-running agent work to stay focused on the original task.

What it delivers

You can give an agent ambitious work and still keep a clear record of what each thread did.

What it does

Central orchestrator

Plans and breaks down work without directly executing commands or editing files.

Thread-and-episode workflow

Launches threads that return structured episodes describing what they accomplished.

Portable onboarding skill

Includes an installable onboarding skill for setting up nac and connecting MCP.

MCP integration

Connects nac to MCP clients so agents can use external tools and services.

Web dashboard and CLI

Provides `nac-web` for starting the dashboard and running upgrades.

How to get it

  1. 1By Hand. Install the latest stable release
    curl -fsSL https://raw.githubusercontent.com/arcee-ai/nac/main/scripts/install.sh | sh
  2. 2The installer puts nac-web in $HOME/.local/bin. Add that directory to your PATH if…
    nac-web
  3. 3Upgrade. Nac ships with a native upgrade function which can be used to install the…
    nac-web upgrade
  4. 4Run
    # Upgrade to the latest nightly release candidate
    nac-web upgrade --pre-release
  5. 5Pick one before you start a session, or configure it later from the dashboard.
    # Recommended: Arcee device-code login → Latest Open & Arcee Trinity models
    nac-web arcee-auth login
    
    # ChatGPT account via OAuth → OpenAI models
    nac-web codex-auth login
    
    # Any catalog provider (DeepSeek, Fireworks, Together, OpenAI, Anthropic, arcee-api, …)
    export OPENAI_API_KEY=...   # or ANTHROPIC_API_KEY, TOGETHER_API_KEY, ARCEE_API_KEY, …
  6. 6Run
    curl -fsSL https://raw.githubusercontent.com/arcee-ai/nac/main/scripts/uninstall.sh | sh

README

nac from Arcee

ArceeDocumentationReleasesArcee Open Model API

CI License

nac is an open-source agent harness for longer, ambitious tasks — experiments, training runs, infrastructure, and prototyping that has to stay aligned with the original intent. It uses a thread-and-episode architecture inspired by slate: a central orchestrator plans and decomposes work but cannot execute commands or edit files; it only launches threads, which return episodes — structured summaries of what they accomplished. Also takes inspiration from nanocode and pi. For the technical write-up, see the nac blog post.

Quickstart

By Hand. Install the latest stable release:

curl -fsSL https://raw.githubusercontent.com/arcee-ai/nac/main/scripts/install.sh | sh

The installer puts nac-web in $HOME/.local/bin. Add that directory to your PATH if needed, then start the dashboard from your project,

nac-web

and navigate to the interface in your browser (default: http://127.0.0.1:3210).

By Agent. Nac provides a portable onboarding skill and MCP integration; paste this into your chosen agent harness to install nac, configure it, and connect the MCP:

Install the nac onboarding skill with curl -fsSL https://raw.githubusercontent.com/arcee-ai/nac/main/scripts/install-skill.sh | sh -s -- --target agents, load nac-onboarding, and walk me through installing nac, selecting Arcee login, ChatGPT Codex login, or an OpenAI-compatible API key, adding MCP servers, and connecting your MCP client to nac.

Upgrade. Nac ships with a native upgrade function which can be used to install the latest stable release of nac.

nac-web upgrade
# Upgrade to the latest nightly release candidate
nac-web upgrade --pre-release

Model Provider Authentication

Pick one before you start a session, or configure it later from the dashboard.

# Recommended: Arcee device-code login → Latest Open & Arcee Trinity models
nac-web arcee-auth login

# ChatGPT account via OAuth → OpenAI models
nac-web codex-auth login

# Any catalog provider (DeepSeek, Fireworks, Together, OpenAI, Anthropic, arcee-api, …)
export OPENAI_API_KEY=...   # or ANTHROPIC_API_KEY, TOGETHER_API_KEY, ARCEE_API_KEY, …

More details can be found in Providers and logins and Model configuration.

Uninstall

curl -fsSL https://raw.githubusercontent.com/arcee-ai/nac/main/scripts/uninstall.sh | sh

Documentation

Contributing

Pull requests are welcome. A CLA-signing bot checks every PR against arcee-ai's CLA; comment I have read the CLA Document and I hereby sign the CLA on your PR to sign it (or recheck to re-run the check).

nac is licensed under Apache 2.0.

Files in the repo

Repository payload15 top-level entries
  • .agents
  • .github
  • crates
  • docs
  • fixtures
  • scripts
  • skills
  • .gitignore
  • Cargo.lock
  • Cargo.toml
  • LICENSE
  • Makefile
  • README.md
  • start_dev.sh
  • start.sh

Discussion (0)

Ask about usage, or say what you built with it

Sign in to join the discussion.

No comments yet. Be the first to say what this is good for.

More harnesses

affaan-m/
ECC
affaan-m/ECCHarnesses

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

258k

The job search that runs on your machine. AI job application framework built on Claude Code: evaluate postings, tailor CVs, write cover letters, prep interviews. Fork it and own it.

42k
ruvnet/rufloHarnesses

🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, RAG integration, and native Claude Code / Codex / Hermes and many more Integrated

72k

Practical patterns, starters & CLI tools for loop engineering with AI coding agents. Design systems that prompt and orchestrate agents (inspired by Addy Osmani and Boris Cherny). Includes loop-audit, loop-init, loop-cost.

11k