Overview
clanker is a self-improving AI agent harness written in Zig 0.16. It runs its tools as sandboxed WebAssembly modules via zwasm, and improves its own source code through a gated loop: the agent proposes an exact-match patch, applies it to a staging copy, verifies it with zig build, zig build test, zig build tools, zig fmt, and lint, and promotes it to the live tree only if all gates pass. The project follows an Everything is a plugin design: tools, models and providers, Web UI views, skills and prompts, and config are plugin surfaces, while the core that remains core (the sandbox policy, the gates, the credential handling) is small on purpose and stays out of any plugin's reach. Most plugins ship as sandboxed WASM modules, so the plugin boundary is also the security boundary. 0.13.0 is the newest published release, and main carries the next one. To build from source, the requirements are Zig 0.16.0, Git, Bash, and patch; the test suite also requires Bun and Python 3. Features include WASM tools, an MCP server, peer notifications and phonebook, mesh clustering, A2A agent cards, goal lifecycle, and a REPL with streaming and slash commands.
Pricing
Free
Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.
Visit WebsitePrices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.
Use Cases
- Teams use clanker to run tools as sandboxed WebAssembly modules.
- Teams use clanker to let the agent improve its own source through a gated loop.
- Teams use clanker to expose tools to MCP clients through a stdio JSON-RPC server.
- Teams use clanker to send messages to other clanker instances and list agent cards.
- Teams use clanker to join, leave, and inspect a TCP cluster of clanker serve processes.
- Teams use clanker to draft, save, and run goal loops.
- Teams use clanker to interact through a REPL with live token output and slash commands.
- Teams use clanker to see per-turn cost and context usage.
Features
Run a self improving agent harness
Control changes with gated self patches
Run tools in sandboxed WASM
Operate agents through a CLI
Access agents through a web UI
Reviews
No reviews yet. Be the first to share your experience!