recollect-memory

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.

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 entries
• Recollect.Knowledge.forget/1 — delete entries
• Recollect.Knowledge.connect/4 — link entries
• Recollect.Search.search/2 — hybrid search
• Recollect.Search.ContextFormatter.format/1 — format for LLM
• Recollect.Pipeline.Ingester.ingest/3 — ingest documents
• Recollect.Pipeline.process/2 — run full pipeline
• Recollect.Maintenance.Decay.run/1 — archive stale entries
• Recollect.Maintenance.Reembed.run/1 — re-embed with new model
• Recollect.Consolidation.run/1 — sleep consolidation
• Recollect.Invalidation.invalidate/3 — weaken deprecated memories
• Recollect.WorkingMemory.push/3 — session-scoped notes
• Recollect.Handoff.create/2 — session continuity
• Recollect.Outcome.good/1 / bad/1 — feedback on recalled memories
• Recollect.Export.export_all/2 / Recollect.Import.import_all/2 — portability

Embedding Providers

• Recollect.Embedding.Local — Default, Bumblebee, 384 dims, no API key
• Recollect.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:

• knowledge.md — Knowledge API, entry types, lifecycle
• pipeline.md — Document ingestion and processing
• search.md — Search, context formatting, completion
• embedding.md — Providers, dimensions, credentials
• maintenance.md — Decay, reembed, consolidation, invalidation
• extension-points.md — All behaviours and extension APIs