What is Lore, in one sentence?
The system of record for enterprise AI agents: what they remembered, what they were allowed to know, what they actually did — with a signed receipt for each.
Which AI tools does it work with?
Capture clients for Claude Code, Codex, Cursor, Windsurf, Cline and Gemini CLI, plus any MCP client connected to Lore; a native MCP server with 13 tools up front and 88 more discovered on demand, served over HTTP; a VS Code extension; a Chrome extension for web conversations; and a CLI. Memory follows the project — not the tool.
Why should anyone trust the receipts?
Receipts are dual-signed (Ed25519 + HMAC-SHA256), hash-chained into an insert-only store, and verifiable offline by a standalone verifier that reports VALID, TAMPERED, or INVALID — no access to our database required.
What happens when an agent lies about what it did?
Lore doesn't take the agent's word for it. After execution, the server independently re-derives authoritative state; contradictions are recorded as state_mismatch, never silently accepted. Destructive operations without a valid authority token simply do not run.
Can we delete memory — provably?
Yes. Cryptographic erasure destroys per-unit keys, producing a receipted, offline-verifiable record of exactly what was destroyed — deletion an auditor can check, not just a soft-delete flag.
Do we have to adopt the whole platform?
No. Most teams start with the self-serve memory and context hub. The execution gate comes when your agents get real application authority, and compliance evidence is an add-on to that — each layer uses the record the previous one built.
How does capture stay under our control?
In-conversation directives control capture per conversation or per user (start/end, on/off, project assignment, titles) — parsed server-side, stripped from stored content, and audited. Directives in assistant or tool turns are ignored, so text discussing a directive never toggles capture.
What infrastructure does it run on?
PostgreSQL with pgvector, Redis, and etcd at the core; SeaweedFS/S3, Neo4j, and an Apache Iceberg lakehouse optional. Deploy SaaS or fully self-hosted. Runs as native services under PM2, or from the signed offline installer. No GPU inference in the serving path.
Which compliance frameworks are covered?
Frameworks are modeled as data on one canonical foundation: SOC 2 today, with HIPAA, GDPR, DORA, the EU AI Act, and others normalizing onto the same entities via versioned crosswalks. Absent evidence is recorded as a gap, never a pass.
Is this actually in production?
Yes. Lore runs in production today, and early access brings the first external tenants onto the record.