| name | recollect-memory |
| description | Use this skill when working with the Recollect Elixir memory engine. Consult for knowledge storage, search, document pipelines, working memory, session handoff, maintenance, and extension points. |
| metadata | {"managed-by":"usage-rules"} |
Overview
Recollect is a pluggable memory engine for Elixir (~> 1.17). It provides three
tiers: working memory (session-scoped), lightweight knowledge (store-embed-search),
and a full document pipeline (chunk, embed, extract, graph).
Quick Reference
Critical: No Top-Level Module
All API lives on submodules:
Recollect.Knowledge.remember/2 — store entriesRecollect.Knowledge.forget/1 — delete entriesRecollect.Knowledge.connect/4 — link entriesRecollect.Search.search/2 — hybrid searchRecollect.Search.ContextFormatter.format/1 — format for LLMRecollect.Pipeline.Ingester.ingest/3 — ingest documentsRecollect.Pipeline.process/2 — run full pipelineRecollect.Maintenance.Decay.run/1 — archive stale entriesRecollect.Maintenance.Reembed.run/1 — re-embed with new modelRecollect.Consolidation.run/1 — sleep consolidationRecollect.Invalidation.invalidate/3 — weaken deprecated memoriesRecollect.WorkingMemory.push/3 — session-scoped notesRecollect.Handoff.create/2 — session continuityRecollect.Outcome.good/1 / bad/1 — feedback on recalled memoriesRecollect.Export.export_all/2 / Recollect.Import.import_all/2 — portability
Embedding Providers
Recollect.Embedding.Local — Default, Bumblebee, 384 dims, no API keyRecollect.Embedding.OpenRouter — API, 1536 dims
No OpenAI or Ollama providers exist.
Database Adapters
Recollect.DatabaseAdapter.Postgres (default, requires pgvector)Recollect.DatabaseAdapter.SQLiteVec (SQLite3 + sqlite-vec)Recollect.DatabaseAdapter.LibSQL
Configuration
Required: config :recollect, :repo. Everything else has defaults.
Use :credentials_fn for API keys, never static config in production.
References
See the reference files for detailed information:
