Overview
TokenDanceCode lets you run AI coding agents on your own machine. Your code never leaves your filesystem, and you control which API keys to use.
Use it to read codebases, fix bugs, write tests, run quality gates, and do diff reviews — all locally. Every agent action is recorded in an auditable JSONL transcript. No cloud platform required, and you are never locked into a single model provider.
What you can do
Local CLI coding agent. Run one command and the agent understands your project, fixes failing tests, or generates code changes. Ideal for fast iteration and automating routine dev tasks.
Bring your own keys. Plug in OpenAI, Anthropic, or the TokenDance Gateway directly. All API calls originate from your machine, and provider selection is controlled via environment variables. No keys? The built-in MockProvider lets you validate installation and run smoke tests anyway.
Connect to AgentHub. The SDK bridges your local runtime into AgentHub v0.4.0, the shared workbench. Get IM-style collaboration, remote approval flows, and a desktop workbench UI. Agent execution events, transcripts, and permission decisions stream into the workbench in real time.
Safe isolated execution. The permission engine gates high-risk operations (file writes, network calls, shell commands) with configurable levels. Subagent and worktree mechanisms isolate changes in dedicated branches that you merge only after review.
Auditable run records. Every session produces a canonical JSONL transcript with the full chain of tool calls, results, and permission decisions. Traceable and replayable.
TokenDanceCode is in early public release. The core runtime is stable; the provider surface and worktree support are expanding. Full roadmap at roadmap-status.
Where to start
- Quickstart — run your first command in 5 minutes
- Providers — configure API keys and model backends
- AgentHub SDK — connect to the shared workbench
- Permissions & Safety — understand the safety boundary
Feedback
Found a bug or have a suggestion? File an issue at github.com/TokenDanceLab/tokendance-site.
Edit this page on GitHub