arcee-ai/nac

View on GitHub ↗

Give AI agents ambitious work without losing the plot. nac is an open-source harness for long-running tasks, using a central orchestrator, threads, and structured episodes to stay aligned with your intent.

137 ★11 forksRustUpdated 18d ago

What you need to know

Open-source agent harness (nac-web dashboard plus MCP integration) for longer, ambitious tasks that must stay aligned with original intent, using a thread-and-episode architecture where a non-executing orchestrator launches threads that return structured episode summaries.

Install

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

Usage

  • •Run nac-web from your project and open the browser dashboard at http://127.0.0.1:3210.
  • •Authenticate a provider first: nac-web arcee-auth login, nac-web codex-auth login, or set an API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, TOGETHER_API_KEY, ARCEE_API_KEY, etc.).
  • •Install via the portable onboarding skill and MCP integration by pasting the provided curl one-liner into your agent harness.
  • •Upgrade with nac-web upgrade (add --pre-release for the latest nightly candidate).

Key features

  • ✓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 that return episode summaries.
  • ✓Keeps long experiments, training runs, and infrastructure work aligned with original intent.
  • ✓Flexible providers: Arcee device-code login, ChatGPT/Codex OAuth login, or OpenAI-compatible API keys.
  • ✓Portable onboarding skill and MCP integration for agent harnesses.

Best for

Long-running, multi-step agentic work (experiments, training runs, infrastructure) that needs intent alignment and structured episode feedback.

Caveats

  • ⚠The README is thin — details live in docs/ (AGENTS.md files, skills, sandbox, model config, credentials, HTTP API).
  • ⚠Agents do exercise real work via threads; sandbox and model-request security are documented in docs/.
  • ⚠Apache 2.0.
Platforms: Local · Web

Reviewed 2026-08-18

Topics

agent-harnessagent-orchestrationagent-skillsagentic-workflowsai-agentsdeveloper-toolsllm-agentslong-running-agentsmcpmodel-context-protocolmulti-agent-systemsrust
Stars
137★
Forks
11
Language
Rust
License
Apache-2.0
Created
2026-03-28
Last push
2026-08-18