Restless

Reads everything. Trusts nothing.

Scouts drug programmes, pulls the evidence, argues with itself, then hands you a licensing verdict. Every claim sourced. Every gap flagged.

NEW Restless 0.5 adds genetics depth from Open Targets and GTEx  →

What the Agents Said

"Response rate comes from 11 patients. The deck calls it a strong signal. I disagree."

@red-team

"Composition of matter expires 2031. Method of use buys two more years. Thin, but licensable."

@ip-agent

"Three competitors reach launch first. By then this is fourth-in-class."

@landscape-agent

"Phase 1 dose escalation complete. No DLTs. Efficacy readout not yet public."

@clinical-agent

"Genetic support for the target is strong. Tissue expression is not where the deck says it is."

@genetics-agent

"Verdict: Pursue. Confidence medium. Two items on the diligence tracker before term sheet."

@orchestrator

How It Works

01 — Scout
Point it at a programme.
Restless pulls trials, papers, patents, filings and pipelines from 14 live data connectors.
02 — Dissect
Nine specialist agents.
Each agent owns a section — mechanism, clinical, IP, market, the lot. No single model sees everything.
03 — Challenge
A rival model family attacks.
Weak claims get rewritten or cut. The red team uses a different model family to prevent self-flattery.
04 — Decide
Ten sections. One verdict.
A scored, sourced dossier and a ranked queue. Not an avalanche of reports — a decision you can act on.
10
dossier sections per programme
9
specialist agents
14
live data connectors
281
programmes screened

Built to Be Trusted

🔗
Sourced or it doesn't ship.
Every claim links back to its evidence. Unverified citations fail the rubric and trigger a retry.
Confidence on every line.
Restless says when it is guessing. High, medium, or low — never silent about uncertainty.
🔒
Your data stays yours.
Data room material runs on local or zero-retention models only. Confidential content never touches shared endpoints.
No lock-in.
Swap any model. The rubrics and knowledge stay on your infrastructure. The IP is in the prompts, not the platform.
Humans hold the pen.
Nothing enters the knowledge base without sign-off. Candidate facts are promoted only after expert review.

Run live

Example programmes

Loading…