Agent-readable docs index: /docs/llms.txt. Full docs in one file: /docs/llms-full.txt. Download /docs/docs.zip to grep all markdown files locally.

Configuration

These are Convex deployment environment variables. None has a secret default.
VariableDefaultWhat it does
KAIBO_AI_URL, KAIBO_AI_SECRETunsetThe Workers AI worker. Without both, kaibo uses deterministic rules and keyword recall only.
KAIBO_AI_DAILY_NEURONS9000Daily Workers AI budget. Past it, no model calls are made and remember falls back to the rules writer. 0 turns the guard off. Cloudflare's free allocation is 10,000 neurons a day.
KAIBO_WRITER_MODELLlama 4 ScoutModel that turns raw events into memories.
KAIBO_JUDGE_MODELLlama 3.3 70BModel that picks memories for a question and finds outdated ones. It reads a lot and writes little, so the larger model is cheap here.
KAIBO_WRITER_BATCH1Most events the writer reads in one call. On Workers AI a model call takes time in proportion to what it writes, so batching saves cost and not time. Raise it for bulk imports to cut the number of calls.
KAIBO_PROVISIONALonNew events become searchable straight away as placeholder memories until the model writer finishes. Set 0 to disable.
KAIBO_ATTEST_SECRETunsetKey for pack attestation. Without it, packs carry no attestation.

What it costs

Every model call is counted against the daily budget. A remember costs roughly 30 to 90 neurons depending on the writer model, and a recall in deep mode about 40. fast recall costs only an embedding. Cloudflare bills Workers AI above the free allocation on paid plans, so set the cap deliberately.

Speed

OperationTypical
remember acknowledgedunder 0.2 s
New event searchableunder 1 s
Event fully structured by the model5 to 7 s, in the background
recall in fast modeabout 0.4 s
recall in auto or deep mode1 to 2 s