Drug Discovery Alpha
Compounds, bioactivity, clinical trials, regulatory signals, and prior-failure records. For agents doing virtual lab work before any wet experiment.
Request accessSciweon
Data infrastructure for AI agents
Structured, agent-ready scientific data and evidence. Records carry their source and, where recorded, a confidence score - so an agent can check a claim instead of taking it on trust.
Domains
Design principles
- Agent-first Designed for machines to consume directly. No human search-UX overhead in the data layer.
- Cross-source intent The goal is that a fact be checkable across independent providers and that disagreement be preserved rather than collapsed. Competing-claim preservation is NOT yet implemented in the pipeline; today a later value can replace an earlier one.
- Provenance Records generally carry the source, timestamp and extraction method they were built from. Coverage is not uniform across every field and is not asserted as complete.
- Confidence recorded Where a confidence score is computed it is returned rather than hidden. It is a recorded score, not a guarantee of correctness.
- Negative evidence Failed trials, retractions, withdrawn drugs, post-market signals - collected as first-class data.
- Dated snapshots Evidence responses report the snapshot date they were served from, so an answer can be pinned to a point in time. No refresh cadence is promised.
For AI agents
Sciweon is not a search engine for humans. The intended user is an AI agent doing real research work - selecting compound leads, scoring trial designs, and avoiding prior-failure repeats.
API and MCP server are in private alpha. No public release date is published. To get on the alpha list: hello@sciweon.com