Live today
- Any surface → a verified graph. OpenAPI, human docs,
llms.txt, Anchor IDLs, raw program source. One command:gecko add <spec-or-docs>. - Provenance on every edge —
extracted/recovered/flaggedon-chain;EXTRACTED→DECLARED→INFERRED→CLAIMED→VERIFIED/REFUTEDon HTTP. find_start— intent → the right starting point, with a dependency-ordered derive plan and an honest no-start below the retrieval floor. See find_start.- simulate → Receipt — a $0 fork simulation before money moves, live-proven twice. See The Receipt.
- Invisible auth — the credential lives in your OS keychain and is injected at call
time. Never in a tool def, never in
mcp.json, never in the model’s context. $0recorded mode — falsify every call offline, from the schema, before you spend a token.- The Scorecard —
gecko report <spec>: a grade plus the specific findings. - Correctness in CI —
gecko test <spec>. - Seven fail-closed security layers, including an AST-enforced never-sign boundary.
- The categorical corpus + an N-confirmed drift detector —
gecko drift <series>. - Hosted MCP — one-click add to Claude Code / Cursor / VS Code.
Building
In progress. The primitives ship today; the scheduled pieces are being built.
- The drift scheduler. The detector works and the series reads back; we are building the cadence that re-simulates on a schedule so the series accrues without you running anything. Until then it accrues only when runs happen.
- Catalog breadth. 4,500 projects are listed and 5 are wired deep; we are widening the auto-comprehend path so more of the catalog resolves without hand-authoring. Non-Anchor generalization is proven once.
- Binding the signing gate to the simulated message hash. Today’s gate is
verdict-based. We are building
evaluate_txso the gate proves the thing being signed is the exact thing that was simulated. - Hosted point-and-simulate, and hosted program surfaces beyond the first.
Exploring
Unbuilt. If one of these is the reason you’d adopt, tell us — it moves up.
- A semantic/vector retrieval tier. Deliberately OFF behind an evidence gate. It flips only on measured lexical recall failure, not fashion — and one measured negative result on embeddings is already on record.
- Cross-customer episodic pooling. Tenancy is local-only until a consent and egress layer exists that we’d be willing to defend.
- Response-level trust. “Is this response sane?” — deliberately downstream of comprehension and verification.
What we’ll never build
So the labels above stay credible:- No signing, no broadcasting, no custody. Gecko simulates and hands a verdict to someone else. Unsigned assembly for simulation is the one documented carve-out, and it is AST-enforced at the sign/send boundary.
- No payment rail, no marketplace. We compose on MCP, x402, and payment catalogs; the provider charges you directly.
- No storage of your API responses, request values, or secrets. The corpus is categorical by construction — a fail-closed allowlist, not a policy you have to trust.
- No browsable provider marketplace. Point us at the surface — the flow is identical for every API.
Who pays
Thegecko-surf engine is free and open-source, forever — Apache-2.0, complete,
self-hostable. Ingest, comprehend, verify, run it yourself.
Developers never pay. The hosted layer is what an API provider buys: a flat price per
API, no cut of an API call, no funds held. See For API providers.
Tell us what to comprehend next
Which API did your agent call wrong?
The call that looked right and wasn’t — the one the spec described perfectly and the API
still refused, or accepted and did the wrong thing. Open an issue and tell us. Public 👍s
rank what we comprehend next.