Singularmem

Singularmem

Singularmem is a local-first persistent memory layer for LLM-driven workflows. It stores, indexes, and exposes the artefacts a developer or agent accumulates over time — conversations, files, decisions, embeddings, provenance — and bridges them to any LLM provider through a stable, vendor-neutral interface.

**Status:** v0.16.0, plus unreleased transcript ingestion (v0.17.0) and scoping (v0.18.0) — memory store, hybrid search (Tantivy + USearch), provider adapters, MCP server, TypeScript SDK, bulk transcript ingestion, and hierarchical item scoping. Constitution v0.2.0 ratified 2026-05-15.

Open core

Singularmem ships as **open core**:

• The open components — memory engine, on-disk format, indexes, embedding pipeline, LLM provider adapters, CLI, MCP server, library SDK, and the TypeScript binding — are licensed under Apache-2.0 and live in this repository.
• The proprietary components — the desktop GUI (Flutter), premium visualisations, and cross-device sync — are sold under a separate commercial license to sustain development.

The boundary between the two is a [constitutional matter](.specify/memory/constitution.md#open--closed-split), not a product-management one. The constitution's Principle III.a is a **one-way ratchet**: features may move from proprietary to open, never the reverse.

Quickstart

# Make every past Claude Code session searchable
singularmem ingest-transcript            # defaults to ~/.claude/projects
singularmem ingest-transcript --project "$PWD"   # only sessions from this repo

# Index a source tree (honours .gitignore; re-runs only pick up changes)
singularmem ingest-dir .

singularmem search "why did we pick tantivy"
singularmem retrieve --adapter claude "release process"

# Every item gets a default scope (claude-code/ or files/);
# use it to narrow a search or see what's in the store
singularmem search --scope claude-code/singularmem "why tantivy"
singul