v0.3.0 — Hierarchical search, plugin install, desktop app

Persistent memory for
AI coding agents

Install the plugin, that's it. Your agents automatically save decisions, search before implementing, and build project knowledge across sessions. Local-first, encrypted at rest, fully yours.

terminal
# 1. Install
$ brew install Rememora/tap/rememora

# 2. Add as a Claude Code plugin
$ claude plugin marketplace add Rememora/rememora
$ claude plugin install rememora@rememora

# 3. Register a project
$ rememora project add myapp --path . \
    --description "My awesome app"

# Done. No manual commands needed.
# Everything is automatic from here.

or cargo install --path . from source • rememora setup --apply for Codex & Gemini CLI

Works with every agent that has Bash access

0ms
Startup Time
0MB
Binary Size
0
External Dependencies
0
CLI Commands

Your AI agents have amnesia

Every session starts from scratch. Switch agents mid-task and context is lost. Come back after a week — your agent has forgotten everything.

Without Rememora

Agents lose all context between sessions. You manually re-explain decisions, repeat yourself, and watch agents make mistakes you already corrected.

With Rememora

Every agent inherits full project history. Decisions, patterns, gotchas, and working state carry over seamlessly — even between different agents.

Browse your memories visually

A native desktop app built with Tauri — dashboard, browser, and full-text search across all your memories.

Rememora Desktop — Dashboard view showing memory stats, category distribution, recent memories and sessions
Browse view with project and category filters

Filter by project & category

Search view with BM25 results

Full-text search with BM25 ranking

Everything your agents need to remember

A complete memory system designed from the ground up for multi-agent workflows.

🔀

Cross-Agent Transfer

Hand off working context between Claude Code, Codex, Gemini CLI, or any agent. Session chains maintain full continuity.

🔎

BM25 + Vector Search

Full-text search via SQLite FTS5 with optional vector similarity. Hybrid RRF merges both for best results.

🤖

Autonomous Curation

LLM-powered extraction from session transcripts. Haiku signal gate + Sonnet AUDN curator. Zero manual work.

📈

Hotness Scoring

Sigmoid-based frequency scoring with exponential time decay. Important + frequently-used memories surface first.

🔒

Encrypted & Local

SQLCipher encryption at rest. Keychain integration. Everything stays on your machine — no cloud, no telemetry.

Memory Evolution

Automatic consolidation of stale and duplicate memories. BM25 clustering + LLM-driven merge, supersede, and pruning.

Six types of knowledge, structured

Memories are categorized for optimal retrieval and context assembly.

🛠 Preference
📚 Entity
Decision
📅 Event
🔎 Case
🔄 Pattern

Preference

"Prefers Zustand over Redux"
"Uses pnpm, not npm"

Entity

"Stripe API uses idempotency keys"
"Auth service at src/auth/"

Decision

"Chose expo-router over React Navigation"
"SQLite over Postgres for CLI"

Event

"v2.0 shipped 2026-03-01"
"Merge freeze starts Friday"

Case

"iOS build fails with Hermes + RN 0.76"
"Fix: disable new arch in pod"

Pattern

"Always run migrations before seeding"
"Use BDD-style test names"

From session to shared memory

Rememora integrates directly into your agent workflow via Bash — no MCP, no server, no infrastructure.

1

Register your project

One command to register. Rememora auto-detects project from your working directory after that.

$ rememora project add myapp --path . --description "Mobile app"
2

Agents save memories

During work, agents save decisions, discoveries, and patterns. Or let autonomous curation extract them automatically.

$ rememora save "expo-secure-store for tokens" \
    --category decision --project myapp
3

Context loads on session start

Every agent gets full project context in one call. Tiered loading keeps token usage minimal.

$ rememora context --auto
4

Transfer between agents

End a session with status and working state. The next agent — any agent — picks up exactly where you left off.

cross-agent-transfer.sh
# Claude Code finishes, hands off
$ rememora session end $SESSION_ID \
    --status transferred \
    --summary "Auth flow 80% done" \
    --working-state "Login UI done.
    Token refresh blocked on
    secure storage decision."

# Switch to Codex — full context
$ rememora context --project myapp
  ## Project: myapp

  ### Active Memories (12)
  ● [decision] Chose Zustand over Redux
  ● [decision] expo-secure-store for tokens
  ● [entity] Stripe API idempotency keys
  ● [case] iOS Hermes fix: disable new arch
  ● [pattern] BDD-style test names
  ...

  ### Last Session (transferred)
  Agent: claude-code | 2h ago
  Summary: Auth flow 80% done
  Working: Token refresh blocked...

# Codex continues with full context
$ rememora session start \
    --agent codex \
    --project myapp \
    --intent "resolve secure storage" \
    --parent $SESSION_ID

Memories that write themselves

Rememora hooks into Claude Code to extract and curate memories from your session transcripts — automatically, incrementally, with audit trail.

Parse

JSONL session transcripts read incrementally via watermarks

Gate

Haiku signal gate: fast YES/NO — does this contain knowledge?

Curate

Sonnet AUDN cycle: Add, Update, Delete, or Noop each memory

Evolve

BM25 clustering + LLM merge/prune removes stale duplicates

autonomous-curation.sh
# Auto-curate all Claude Code sessions
$ rememora curate --auto
  Scanning 14 session files...
  ▸ 3 new transcripts found
  ▸ Signal gate: 2 passed, 1 filtered
  ▸ AUDN results: +4 added, ~1 updated, -0 deleted
  ▸ 5 memories curated in 3.2s

# Preview without applying
$ rememora curate --auto --dry-run

# Evolve: merge stale duplicates
$ rememora evolve --project myapp
  Found 3 clusters of similar memories
  ▸ Merged 2 decision memories
  ▸ Superseded 1 stale pattern
  ▸ Kept 4 memories as-is

Three layers of integration

From raw CLI commands to fully autonomous multi-agent orchestration.

Layer 3: Multi-Agent Orchestration
agent-run, agent-loop, developer/triage agents, atomic locking, git worktrees, GitHub Project board integration
Layer 2: Claude Code Plugin
Hooks (SessionStart, SessionEnd, Stop) • Skills (save, search, init) • Autonomous curation after every turn
Layer 1: CLI Core
save, search, context, session, curate, evolve, extract, project, export, tui • SQLite + FTS5 + WAL • SQLCipher encryption

Tiered Context Loading

L0 — Abstract (~100 tokens): memory map for quick orientation
L1 — Overview (~500 tokens): top memories with detail
L2 — Full content: complete text when needed

Hotness Scoring

sigmoid(log1p(access)) * exp(-age/7d)

Blended 70% importance + 30% hotness. Frequently-accessed, recent, important memories always surface first.

Dispatch issues to agents autonomously

Rememora can watch your GitHub project board, dispatch issues to Claude CLI, and open PRs — with quality gates and retry loops.

🚀

agent-run

Dispatch a single GitHub issue to Claude CLI. Creates isolated git worktree, runs tests, retries on failure, opens PR, updates project board.

$ rememora agent-run \
    --repo owner/repo \
    --issue 42 \
    --retries 3
🔁

agent-loop

Continuous polling of your project board. Auto-dispatches "Ready-For-Dev" issues. Merges "Cherry-Picked" PRs. Configurable poll interval.

$ rememora agent-loop \
    --repo owner/repo \
    --poll 300

Built different, by design

Feature Rememora MCP Memory File-based
Cross-agent support Any agent with Bash MCP-only agents ~ Manual sync
Session transfer Built-in chains No No
Full-text search BM25 + Vector + RRF ~ Basic grep
Autonomous curation LLM signal + AUDN Manual Manual
Memory evolution Merge, supersede, prune No No
Encryption SQLCipher ~ Varies Plaintext
Infrastructure Zero (single SQLite) ~ MCP server Zero
Startup time ~3ms ~ ~100ms+ ~0ms

What's new

v0.3.0

April 2026
  • Hierarchical score propagation — search boosts related memories via URI tree
  • Claude Code plugin install/plugin install rememora@rememora
  • Desktop app — Tauri v2 with dashboard, browser, and search
  • Landing page & docs site — GitHub Pages with full documentation

v0.2.0

April 2026
  • SQLCipher encryption at rest — keychain integration
  • TUI dashboard — interactive terminal browser (ratatui)
  • BDD-style tests — Given-When-Then test structure
  • Autonomous curation — LLM-powered memory extraction from session transcripts

v0.1.0

March 2026
  • Core CLI — save, search, context, sessions, projects
  • BM25 search — FTS5 full-text search, zero dependencies
  • Vector search — optional cosine similarity via sqlite-vec + Candle
  • Agent orchestration — agent-run, agent-loop, GitHub project board integration

View all releases on GitHub →

Give your agents long-term memory

Open source. MIT licensed. Built in Rust.