Start as a researcher

Install and choose a strong model 15 min

curl -LsSf https://raw.githubusercontent.com/lpalbou/AbstractFramework/main/scripts/install.sh | sh

Research workflows make many LLM calls, so pick the strongest model your machine fits in the console’s Models tab, or add a cloud provider under Providers.

Run deep-research 20 min

deep-research investigates a request, runs rounds of adversarial review, keeps a verified source ledger and exports Markdown, PDF and DOCX. Its inputs are the request, a viewpoint, and an effort of quick, standard or thorough. It serves the plain chat interface, so AbstractCode can run it directly:

abstractcode login --token <admin token>   # once; the installer’s summary prints this line
abstractcode --workflow deep-research

Or open the Observer, choose Launch and run it once. Its research agents use a read-only evidence allowlist (web search, fetch, skim, read file); only the export step writes files.

Follow every step 10 min

In the Observer, open the run and replay its ledger: each investigation round, every search and fetch, the reviewer’s guidance and the export manifest with file paths and hashes. Nothing is summarized away.

Generate hypotheses with co-scientist 15 min

co-scientist grounds itself in the literature through the deep-research investigation flows, then cycles through generation, reflection, Elo-ranked pairwise debate and evolution, and ends with a reviewed research overview. Launch it from the Observer.