IndraDB / documentation

Concepts

A record of decisions, with branches and history.

Entries

An entry holds reasoning text and an embedding for search. Each entry has a content hash. An entry can also have a custom ID.

Commits

A commit is a snapshot with parent references and a message. Every change makes a commit, so the full history stays. Compare any two points to see what changed.

Branches

A branch gives an experiment its own line of history. Create a branch before you try another approach. Compare it with main. Switch back when you finish.

indra branch experiment
indra checkout experiment
indra create "What if we used microservices instead..."
indra diff main
indra checkout main

Semantic search

Search uses embeddings to find related meaning, not exact words. For example, a query about a database choice can find a saved PostgreSQL recommendation.

Local embeddings are on by default. The model issentence-transformers/all-MiniLM-L6-v2. It needs no API key. Treat results as candidates to review, not as proof that a decision is correct.

The .indra file

The database is one portable file. It stores entries, commits, trees, an index, and branch references. Content addressing stores repeated content once. Objects use zstd compression.

Copy the file to back it up. Use the sync commands to send it to the web app.

Next steps

Connect an agent or read the web platform guide.