Practical overview
When it fits
Use LMC-5 when building a long-running LLM agent that needs to recover selected context across sessions, and when you want to choose between a minimal local SQLite implementation and a richer PostgreSQL/pgvector reference backend.
Who it is for
- Developers building coding agents that need continuity across sessions or context windows.
- Builders of personal-assistant agents and local CLI workflows that need persistent memory.
- Agent developers choosing between a minimal SQLite implementation and a PostgreSQL/pgvector backend.
Documented uses
- Store raw interaction events and curate selected information into durable memories.
- Retrieve relevant information through lexical, vector and, in the production implementation, graph-based recall.
- Represent fact changes and relationships rather than treating each stored item as an isolated embedding.
- Design redaction policies for agent memory records.
Requirements & limits
The repository is public and specifies the AGPL-3.0 license. The README documents a minimal implementation install using `pip install -e .` and a Python demo.
- The minimal implementation uses SQLite with no additional dependencies and combines FTS5 lexical search with portable cosine similarity.
- The production reference implementation uses PostgreSQL and pgvector, with a three-tier recall cascade and multiple recall channels.
- The production backend requires PostgreSQL and at least one embedder API key.
- Production relation-graph recall depends on scheduling the hippocampus relation-build phase; the README recommends adding
night_dreamto a daily cron schedule. - Some automated outputs remain proposals or require review, including applying fact supersession and making archive, delete, merge or demotion decisions.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the minimal SQLite implementation
Install LMC-5 and initialize a local SQLite database for the minimal implementation.
Before you start
- Python 3
- The LMC-5 repository cloned locally
Steps
Create and activate a Python virtual environment.
Install the project with the command below.
pip install -e .Initialize a database with
lmc5 init --db demo.sqlite.Run the documented Python demo with
PYTHONPATH=src python examples/demo.py.
Expected outcome
A local SQLite database is initialized and the project demo can be run using the documented command.
Guide limitations
- This covers setup and the documented demo only; it does not configure an agent integration.
- The minimal implementation is described as suited to prototypes, demos, offline use and fewer than 5,000 vectors.
Prepare the production backend prerequisites
Identify the documented infrastructure and credential prerequisites for the PostgreSQL/pgvector implementation.
Before you start
- A PostgreSQL instance
- At least one embedder API key
Steps
Review the production backend README and its environment example.
Provide the required PostgreSQL instance and an embedder API key before proceeding with the production implementation.
Expected outcome
The stated infrastructure and credential prerequisites for the production backend are identified.
Guide limitations
- This is a prerequisites guide, not a deployment or integration procedure.
- Production setup details are in the backend README and environment example.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 12 October 2026
- Original website title
- wuxuyun0606-collab/lmc-5 · Official homepage title · retained in its source language