Releases: AreevAI/dejadb
Releases · AreevAI/dejadb
Release list
DejaDB 1.0.0
The first public release of DejaDB — the embedded memory engine for AI agents. Memory that doesn't rot: immutable, content-addressed grains in a per-file Turso database, queried with CAL (the Context Assembly Language), rendered to model-ready context in-process. Recall in microseconds — fast enough to run inside a real-time voice agent's turn.
Install
cargo install dejadb # the `deja` CLIRust crates are on crates.io. Python (PyPI) and Node (npm) bindings publish from this release.
Highlights
- Doesn't rot, measured — byte-identical re-writes collapse to one grain; updates are supersessions (1 current value, 0 stale, full history kept); 100% provenance. No LLM in the loop.
- CAL, with no bulk delete —
DELETE/DROParen't tokens; the one destructive statement is a gated single-grainFORGET. - Model-native — built-in MCP server: one line gives any MCP client persistent memory.
- Hybrid recall — structural + BM25 + vector, RRF-fused; multilingual.
- Yours — AES-256-GCM encryption at rest (Argon2id), git-style op-log sync, forks with explicit merges, crypto-erasure.
- A format you keep — the
.mgformat is documented and OMS-conformant, anddeja migrateimports from mem0, Zep/Graphiti, Letta, LangMem, Basic Memory, or generic JSONL.
Full changelog: CHANGELOG.md · Docs & benchmarks in the repo.
Licensed under MIT OR Apache-2.0.