Surfil
Surfil/Products/Pilot
// surfil pilot · consumerpartly built

Ask your own captured knowledge, on your machine.

The answer is often something you or your agent already worked out - a decision, a convention, a gotcha - and re-deriving it costs another expensive round trip.

surfil · one interception point
Claude CodeCursorCodex · Copilot
◈ interceptor
1. security2. cost3. quality4. observability5. memory
≋ one signed Weave spine
Pilot is partly built. Some of what follows ships on a linked device today and the rest does not. Your dashboard marks each capability, and nothing shows a number it has not measured.
What it does

Ask what you already know

Pilot ranks the facts Mind holds and hands back the matching ones with their provenance. It RETRIEVES, so it cannot invent an answer, and it runs entirely on your machine: no model is called, no API is used, nothing is uploaded. Facts sync between your own machines end-to-end encrypted, so any linked machine can answer from all of them.

Retrieval, not generation - Every line is a fact you captured, so a hallucinated answer is not possible.
Answers across your machines - Facts sync sealed; a machine that restored your key answers from all of them.
Nothing leaves the device - Ranking is arithmetic over your own facts. No model call, no API, no cost.
Same provenance model - Every fact carries the code-hash and staleness checks of local Weave.

Proof, not promises

Pilot produces one metric type on the shared spine: answers with provenance, never generated. Every paid output is signed (Ed25519) and verifiable offline with no account. Zero-trace: your source never leaves the device.

$ surfil verify rcpt_8f2a
✓ VALID (offline · epoch 7)
In your terminal

Pilot, as you'd actually see it

claude-code - pilot
$ surfil pilot ask "what signing algorithm do releases use?"
From your own knowledge (1 match(es)):
· surfil uses Ed25519 for release manifests
source: note · relevance 45%
$ surfil pilot ask "which queue library did we pick?"
Nothing you have captured answers that. Pilot returns nothing rather than guessing.
How it works

How Pilot does it

1You capture facts with `surfil mind remember`, or your agent does as it works.
2`surfil mind sync` seals each fact on the device and syncs it between your machines.
3`surfil pilot ask` ranks your facts by how much of the question they cover.
4Every answer cites the fact and its source, so you can check it.
5If nothing you captured covers the question, Pilot says so rather than guessing.
Use cases

Where Pilot earns its place

Stop re-deriving

The decision you made last month answers the question instead of another agent round trip.

Move between machines

A laptop that restored your recovery code answers from everything your desktop learned.

No-hallucination answers

Nothing on record beats a confident wrong answer inside a production codebase.

Claude CodeCursorCodexCopilotMCP agents
FAQ

Questions developers ask first

Does Pilot upload my code or my notes?

No. Ranking runs on your machine. Facts sync between your own machines encrypted end-to-end, and the server holds only ciphertext it has no key for.

Does it call a model?

No. It ranks the text you captured - stemming, stopwords and term rarity. That is also why it costs nothing to run and cannot invent an answer.

What if nothing covers the question?

Pilot says so plainly instead of inventing something.

How is this different from Mind?

Mind stores and syncs the facts; Pilot is how you ask them a question.

Is there a shared team knowledge base?

Not today. Facts are per-account and sync across that account's machines. A team-shared knowledge base is a later step, and it needs a key-sharing design we have not shipped.

Works well with

One spine - products compound

Add Pilot to your agents.

Core installs with Starter; Pilot plugs into the same interception point - no second layer, no new setup.