The harness is the lever.

Get started Read the handbook

$curl -fsSL https://get.veyyon.dev | sh

then vey in any repo to launch

Plan mode

Ground, propose, execute.

Veyyon reads your code and lays out a plan before it touches anything. Tools stay read-only until you approve — so nothing mutates on a guess, and you always see the shape of the work first.

Edits that land

Anchored to content, not line numbers.

Edits attach to what the code says, then get checked before they apply — so far fewer land malformed. When a patch drifts, the harness repairs it instead of corrupting your file.

Models

Three models, chosen by you.

The model you talk to, the one that runs your subagents, and the one that compacts context — each set on its own. No opaque per-role maze, and default is never a model.

model: openai/gpt-5.6-sol
subagent.model: anthropic/claude-haiku-4-5
compaction.model: openai/gpt-5

Cockpit

Watch every agent at once.

A live multi-agent view: a model per subagent, drill into any one, and follow the work as it happens. Fork, clone, and label threads to explore a path without losing the trunk.

Extend

MCP, skills, hooks.

Bring MCP servers, authored skills, and lifecycle hooks — plus browser, web search, and headless exec for CI. The harness is yours to shape.

~/acme/apiopenai/gpt-5.6-sol

› add rate limiting to the upload endpoint ◆ plan ├─ read routes/upload.ts ├─ grep "rateLimit" ok 3 matches └─ edit routes/upload.ts ok +18 −2 running tests … ok 42 passed esc to interrupt ? shortcuts / commands

Get started