Setup guide
Get RE-call running in seven steps.
Written for someone who has never installed it. Every command, what should come back, and what to do when something else does.
What you are installing
A memory that would rather say nothing than guess.
RE-call stores your notes, decisions and project history, and searches them by meaning rather than by keyword. That much is ordinary.
The part that is not ordinary: when nothing it holds actually answers your question, it says so. An ordinary vector search always returns its closest match, because closest is all it knows how to compute. Closest is not the same as correct.
Asked something it has no answer to
[ABSTAIN GAP DEGRADED:INDEX_NOT_READY] query='how do we handle penguins on Mars?' reason: no hit above the calibrated confidence threshold (probable corpus gap) low_confidence conf=0.30 cos=0.458 rate-limits.md 'Clients may issue 100 requests...'
Output from step 7 of this guide, with two diagnostic lines removed to fit. The note it found is still shown, marked low confidence: you are told what was there and why it was not good enough, rather than being handed it as an answer.
- You need
- Python 3.11+3.12 to 3.14 also supported
- And
- Docker DesktopRuns the database for you
- Your data
- Stays localThe default embedder runs on your machine
- Licence
- Apache 2.0Free for commercial use
The moving parts
Three things, and only one of them is RE-call.
Most of the confusion in a first install comes from not knowing which piece a command is talking to. There are only three.
| Piece | What it does | You touch it |
|---|---|---|
| A PostgreSQL database | Holds your memory: the text of every note, plus a numeric fingerprint of its meaning. The pgvector extension is what makes searching those fingerprints fast. | Once, in step 1. It then runs in the background. |
| The RE-call package | Reads your markdown files, turns them into those fingerprints, and decides at query time whether anything it found is trustworthy enough to answer from. | Steps 2 to 7. This is the part you actually use. |
| The MCP server | Optional. Exposes your memory to Claude Code as a tool it can call, so it checks what you have already decided before it answers. | Step 7, and only if you want it. |
The rest of this site
Four pages you will want after the install, and not before.
- Your memory What makes a good memory file, how to organise a corpus that keeps growing, and what not to put in it. →
- Claude Code Every tool the server exposes, and how to get Claude checking memory before it acts rather than after. →
- Calibration Measure, on your own corpus, how confident a result must be before RE-call answers from it. →
- Troubleshooting The errors a first install produces most often, what each one actually means, and the fix. →
Start at step one.
Two things to install, one command each, and a checkpoint after every step so you always know whether it worked.
Start the install