Documentation

Current state (v0.1)

Live: wallet sign-in (sign a free message with a Solana wallet), research workspace, real paper search via OpenAlex, launch draft builder with validation, watchlist, wallet connect (Phantom-compatible) and cluster switching.

Not live yet: AI analysis of research runs, token creation, pools and trading. Runs are saved with status "queued"; drafts are saved with status "draft".

Accounts & privacy

Anyone can browse. Saving research runs, launch drafts and watchlist entries requires an account.

Every saved record is owner-only: access rules on the database ensure only you can read, change or delete your rows.

Clusters

biocode defaults to Solana devnet. Switching to mainnet disconnects your wallet and clears all cluster-dependent data in the app, then reloads drafts and watchlist for the new cluster.

Research

A run combines a question, papers found live in Europe PMC, Crossref and arXiv, pasted text or a private PDF, a depth and stance, and up to 12 specialist roles: Literature mapper, Hypothesis builder, Methods critic, Statistics reviewer, Reproducibility reviewer, Data interpreter, Novelty scout, Translational researcher, Ethics reviewer, Funding planner, Peer-review challenger and Synthesis editor.

Jev Latest makes the typed decisions: it classifies the domain, scores how relevant each source is (low-relevance sources are left out), rates how well each role fits, and estimates whether each cited excerpt supports its claim. Claude Sonnet 5.5 writes each role's findings, hypotheses, dissent and limitations, and must cite sources by quoting them word for word.

A citation is marked verified only when its quoted excerpt is found word for word in the stored source text. Sources are labeled abstract only, metadata only, pasted text or full-text PDF. Verified does not mean the claim is true.

Each run moves step by step while you have it open, and you can cancel it or retry a failed step. Download a Markdown or JSON reproducibility pack with sources, the claim ledger, the checklist, and the model and prompt version for each step.

Current status: paper search, Jev and Claude Sonnet 5.5 (through Anthropic directly, since Lovable AI does not offer that model) all answered live test requests.

Launchpad

Four validated steps: basics (title, symbol, category), thesis and evidence links, treasury address and token settings, then review.

Next: launching will create the token and a Meteora pool on the selected cluster.

Markets & portfolio

No prices or balances are simulated. Markets and holdings appear only from real on-chain data once the Meteora integration ships.