A research prototype for small, testable AI help

When an AI gets stuck,
get one answer. Test it.

Needle gives a specialist only the minimum synthetic context it needs, checks the returned answer with independent tests, and shows why the route won. You can inspect every disclosed input, provider output, verification check, and cost.

SYNTHETIC TASKSMINIMUM DISCLOSUREINDEPENDENT CHECKSDOWNLOADABLE TRACE

Research simulation—not a real marketplace.Mock runs have $0 in API charges. Live mode sends synthetic packets to OpenRouter; its configured API reservation limit is shown before you can confirm a run. Payments, wallets, and irreversible tools are disabled.

transaction envelope

01 describe the missing capability
02 disclose minimum packet
03 request bounded interventions
04 test each returned answer
05 choose verified net value
06 settle simulated credits
07 export redacted trace

Selected experiment

Concurrent billing retry

Executable concurrency verification over a bounded synthetic fix.

Run modeRouting policy

Resolved modeSIMULATION · $0 API COST

This run uses deterministic local fixtures. It does not call OpenRouter, so its actual API cost is $0.

Simulated marketplace credit cap

0.160 cr

Internal “cr” units limit specialist purchases. They are not API dollars or OpenRouter account credit.Independent market auditRequires independent specialist checks even if the low-cost answer passes. This opens the market; “Collect all eligible bids” controls whether it stops early.Collect all eligible bidsAfter the market opens, off enables policy-compliant early settlement.Optional frontier comparisonIn Live mode this can add an OpenRouter API charge; in Simulation it is a fixture. Accounted separately.

Run state: idle

Actual server events

Run progress

READY

No run startedStart a bounded mock experiment to see streamed progress.

When an AI gets stuck,get one answer. Test it.

Concurrent billing retry

Run progress

When an AI gets stuck,
get one answer. Test it.