Git for what your
AI agents know.
Deterministic, verifiable, mergeable persistent memory layer for Cursor, Claude Code, Antigravity, OpenCode, Codex & Gemini.
Seamless Ecosystem
One Memory Layer. Works with all your coding agents.
Lore integrates via Model Context Protocol (MCP). Switch agents freely without losing architectural context or re-teaching what was already discovered.
Cursor
Native MCPIDE agent with inline edits & chat
lore setup cursor Claude Code
Auto ConfigCLI terminal agent by Anthropic
lore setup claude Google Antigravity
Full SupportAutonomous coding CLI & subagents
lore setup agy OpenCode
Auto IngestOpen source multi-agent IDE engine
lore setup opencode Codex CLI
Direct SyncTerminal automation & agent workflows
lore setup codex Gemini CLI
Stdio MCPGoogle DeepMind CLI agent suite
lore setup gemini The Trust Problem
Stop feeding your AI agents append-only vector junk.
Traditional agent memory stores everything in vector databases. But vector search cannot tell if code has drifted, cannot resolve contradictory facts, and cannot undo mistakes.
Without Lore: Blind Vector Memory
- • Dead facts keep living: A file changed 3 weeks ago; the vector database still injects old signatures, leading to broken hallucinations.
- • Contradictions are both "true": Agent A and Agent B learn opposing rules in parallel sessions; search returns both and the model is confused.
- • Context poisoning is irreversible: A hallucinated session pollutes the memory index forever; there is no rollback or undo mechanism.
- • Massive token waste: Because memory cannot be verified, agents repeatedly re-read the entire repository on every prompt.
With Lore: Verifiable Git-Like Engine
- • Automatic drift detection: SHA-256 evidence hashing invalidates claims the instant source code changes; stale facts drop out automatically.
- • Deterministic 3-Way DAG merge: Authority ladder (T0 test > T1 AST > T2 human > T4 inference) resolves conflicts without guessing.
- • Atomic undo with
lore revert: Poisoned sessions are rolled back with an inverse commit, preserving complete audit history. - • 72.3% token savings: Injects precisely the top 10 relevant verified facts instead of loading whole directories into the context window.
Under the Hood
Engineered for absolute accuracy.
Lore behaves like a cache of verified reality, backed by an immutable SQLite DAG and bi-temporal timeline.
Deterministic AST Ingestion
Extracts code invariants, exported routes, dependencies, and environment variables directly from AST parsers. Zero LLM calls, zero token costs, and 0% hallucination risk.
lore ingest # parses AST, env & routes deterministic Cryptographic Drift Detection
Every stored claim is tied to a content hash of its source file. On read, hashes are recomputed in milliseconds. When code changes, the fact turns stale immediately.
lore verify --strict # exit 3 if stale deterministic 3-Way DAG Merge & Authority
Parallel agent sessions branch and merge like Git. When two agents disagree, Lore applies a strict hierarchy: T0 Test > T1 AST File > T2 Human Decision > T4 Inference.
lore merge agent/claude # DAG merge deterministic Solo Permissive Trust Mode
Solo developers don't want to approve drafts manually after every prompt. Permissive mode automatically promotes verified inference while maintaining full audit blame.
lore init --full --solo # zero-friction deterministic Empirical Benchmarks
Evidence, not marketing claims.
Lore ships with executable evaluation suites (npm run eval). Numbers measured on real code fixtures without LLM judge bias.
Compared to full-repo context dumps across 20 replayed benchmark sessions.
Zero hallucinations. Claims are derived directly from source code AST and hashes.
Inverse commit undoes bad sessions without breaking historical blame graphs.
Powered by local SQLite with content-addressable indexing. No external API lag.
Interactive Tour
60 Seconds with the Lore CLI.
Click through the lifecycle of a claim: initialization, AST ingestion, scored search, automatic drift detection, audit blame, and atomic revert.
[lore] Initialising .lore/ database (SQLite v3)...
[lore] Generating lore.ontology.yml...
[lore] Detecting stack: Node.js 22 + TypeScript + Tailwind
[lore] Solo permissive mode enabled (draft gating off).
[lore] MCP server configuration detected:
✓ Antigravity CLI (.agents/mcp_config.json)
✓ Claude Code (.mcp.json)
✓ Cursor (.cursor/mcp.json)
[lore] Ready. 0 active claims. One-Line Setup
Ready in 3 simple commands.
Install globally, initialize your repo, and wire up your favorite AI coding agent via standard MCP in seconds.
Install Lore Globally
Requires Node.js 22 or newer. Installs the core CLI and MCP server executable.
npm i -g @529studio/lore Initialize Repository
Detects your codebase, derives AST claims, and activates solo permissive mode.
lore init --full --solo Connect Your Agent
Automatically writes MCP configurations for Cursor, Claude, AGY, Codex, etc.
lore setup cursor