RecallNest

Shared Memory Layer for Claude Code, Codex, and Gemini CLI

One memory. Three terminals. Context that survives across windows.

A local-first memory system backed by LanceDB that turns scattered conversation history into reusable knowledge — shared across your coding agents, recalled automatically.

English | 简体中文 | Roadmap

Why RecallNest?

Coding agents forget everything between windows. Your context — project configs, debugging decisions, entity mappings — is scattered across Claude Code, Codex, and Gemini CLI with no shared memory.

RecallNest solves this: a single LanceDB-backed memory layer that all three terminals read and write. Context stored in one window is auto-recalled in another. Sessions checkpoint on exit and resume on start. Memory decays, evolves, and self-organizes — not just raw log storage.

Benchmark: LongMemEval (ICLR 2025)

Evaluated on 500 questions across 6 memory abilities (methodology):

Wins or ties in all 6 categories, with no regression. The hybrid retrieval pipeline (BM25 + vector + recency + RIF dedup) surfaces 12.2% more relevant context than vector-only search.

Quick Start

Option A: Claude Code Plugin (recommended)

/plugin marketplace add AliceLJY/recallnest
/plugin install recallnest@AliceLJY

RecallNest starts automatically with Claude Code. No manual MCP config needed.

Requires: Bun (recommended) or Node.js 18+. Dependencies install on first start.

Option B: npm install

npx recallnest --help          # run directly
# or
npm install -g recallnest      # install globally
recallnest doctor

Works with Node.js 18+ (via tsx) or Bun. No git clone needed.

Option C: Manual setup

git clone https://github.com/AliceLJY/recallnest.git
cd recallnest
bun install
cp config.json.example config.json
cp .env.example .env
# Edit .env → add your JINA_API_KEY

Start the server

bun run api
# → RecallNest API running at http://localhost:4318

Try it

# Store a memory
curl -X POST http://localhost:4318/v1/store \
  -H "Content-Type: application/json" \
  -d '{"text": "User prefers dark mode", "category": "preferences"}'

# Recall memories
curl -X POST http://localhost:4318/v1/recall \
  -H "Content-Type: application/json" \
  -d '{"query": "user preferences"}'

# Check stats
curl http://localhost:4318/v1/stats

Connect your terminals

bash integrations/claude-code/setup.sh
bash integrations/gemini-cli/setup.sh
bash integrations/codex/setup.sh

Each script installs MCP access and managed continuity rules, so resume_context fires automatically in fresh windows.

Index existing conversations

bun run src/cli.ts ingest --source all
bun run seed:continuity
bun run src/cli.ts doctor

Web UI

bun run src/ui-server.ts
# → http://localhost:4317

Core Capabilities

Access & Setup

Recall & Continuity

Memory Lifecycle & Governance

Reasoning & Structure

Visibility & Operations

New in v2.1: Philosophy-Informed Memory

v2.0 built the operational memory platform; v2.1 added philosophy-informed memory behavior.

Five upgrades derived from 9 research dimensions in philosophy of memory, each mapped to concrete engineering:

• Emotion-Aware Decay (Affective Memory Theory) — Memories with strong emotional content decay 20-30% slower. Keyword-based emotion detection computes salience (mnemonic significance), which feeds into the Weibull half-life formula and a rebalanced 4-factor evolution score. Zero LLM cost.

• Memory Ethics Layer (Right to Be Forgotten / GDPR Art. 17) — Four privacy tiers (ephemeral / private / durable / shared). Cascade forgetting engine that propagates deletion through KG triples, evolution chains, pin assets, and briefs. Full audit trail. forget_memory MCP tool for agent-driven deletion.

• Autobiographical Narrative (Narrative Identity Theory / Conway's 3-layer model) — Memories are tagged with lifePeriod → generalEvent → specificEvent hierarchy, orthogonal to existing 6 categories. Retrieval pulls narrative siblings. Context rendering groups by life period. Rule-based tagger with EN+CN support.

• Constructive Retrieval (Simulation Theory / Michaelian) — Instead of returning raw stored text, RecallNest now reconstructs context from an expanded candidate set: KG neighbors + evolution chains + cluster members + narrative siblings. Source-map grounded coverage replaces lexical overlap. Contradictions are detected and flagged.

• Predictive Prospective Memory (Mental Time Travel / Tulving) — Heuristic prediction engine that surfaces "you might need this" reminders from behavioral signals: stale checkpoint open loops, corrected workflow observations, high-frequency dormant memories, and uncovered query topics. Zero LLM cost. Auto-expire in 7 days if unaccepted.

New in v2.2: Retrieval Quality Hardening

v2.1 added philosophy-informed behavior; v2.2 closes the last three engine-layer gaps identified by a frontier research scan (ACC, PI-LLM, TSM).

• Memory Confidence Meta-tags (ACC / Dual-Process UQ) — Each memory now carries structured ConfidenceMetadata (score, reliability tier: direct / inferred / hearsay). Auto-assigned from source on write (manual = 0.9, agent = 0.7, conversation_import = 0.5). Retrieval scores are weighted by confidence. resume_context tags low-confidence items with [低置信].

• Interference Detection + Active Forgetting Gate (PI-LLM / SleepGate) — Semantic cluster detection identifies groups of near-duplicate memories competing for retrieval. Enhanced RIF keeps only top-K (default 3) per cluster; extras are demoted 50% instead of removed. Write-time pre-warning: when a scope accumulates ≥5 high-similarity active memories, the weakest is flagged pending_review. data_checkup reports interference density.

• Temporal Validity Windows (TSM / TiMem / Zep) — store_memory accepts validUntil (expiration) and eventTime (when the event actually happened). search_memory supports validAt (point-in-time query) and includeExpired (demote 80% instead of hide). Auto-GC applies 2× decay acceleration to expired memories.

New in v2.3: Connector Ecosystem + Source Health

v2.2 hardened retrieval quality; v2.3 opens RecallNest to external data sources with a standard connector framework and operational health monitoring.

• Connector-v1 Standard (GB-2) — A JSON format (ConnectorOutputV1) that any external script can produce. Obsidian vaults, emails, RSS feeds, log files — normalize once, ingest through the full dedup/embed/extract pipeline. See docs/connector-spec.md for the specification and connectors/examples/ for adapter skeletons (email, logs, RSS).

• Obsidian Vault Ingestion (GB-1) — First-party Obsidian connector: scans .md files, extracts frontmatter + wikilinks, maps folder structure to tags. One command: lm ingest --obsidian /path/to/vault.

• Source Health Monitoring (GB-3) — Every connector ingest writes a heartbeat to data/source-heartbeat.json. data_checkup flags stale sources (>7d warning, >30d error). doctor --ci shows a per-source heartbeat summary with human-readable age.

Architecture

┌──────────────────────────────────────────────────────────┐
│                     Client Layer                          │
├──────────┬──────────┬──────────┬──────────────────────────┤
│ Claude   │ Gemini   │ Codex    │ Custom Agents / curl     │
│ Code     │ CLI      │          │                          │
└────┬─────┴────┬─────┴────┬─────┴──────┬──────────────────┘
     │          │          │            │
     └──── MCP (stdio) ───┘     HTTP API (port 4318)
                │                       │
                ▼                       ▼
┌──────────────────────────────────────────────────────────┐
│                   Integration Layer                       │
│  ┌─────────────────────┐  ┌────────────────────────────┐ │
│  │  MCP Server         │  │  HTTP API Server           │ │
│  │  41 tools           │  │  21 endpoints              │ │
│  └─────────┬───────────┘  └──────────┬─────────────────┘ │
└────────────┼─────────────────────────┼───────────────────┘
             └──────────┬──────────────┘
                        ▼
┌──────────────────────────────────────────────────────────┐
│                     Core Engine                           │
│                                                           │
│  ┌────────────┐  ┌────────────┐  ┌─────────────────────┐ │
│  │ Retriever  │  │ Classifier │  │ Context Composer     │ │
│  │ (vector +  │  │ (6 cats)   │  │ (resume_context)     │ │
│  │ BM25 + RRF)│  │            │  │                      │ │
│  └────────────┘  └────────────┘  └──────────────────────┘ │
│  ┌────────────┐  ┌────────────┐  ┌─────────────────────┐ │
│  │ Decay      │  │ Conflict   │  │ Capture Engine       │ │
│  │ Engine     │  │ Engine     │  │ (evidence → durable) │ │
│  │ (Weibull)  │  │ (audit +   │  │                      │ │
│  │            │  │  merge)    │  │                      │ │
│  └────────────┘  └────────────┘  └──────────────────────┘ │
└──────────────────────────┬───────────────────────────────┘
                           ▼
┌──────────────────────────────────────────────────────────┐
│                    Storage Layer                          │
│  ┌─────────────────────┐  ┌────────────────────────────┐ │
│  │ LanceDB             │  │ Jina Embeddings v5         │ │
│  │ (vector + columnar) │  │ (1024-dim, task-aware)     │ │
│  └─────────────────────┘  └────────────────────────────┘ │
└──────────────────────────────────────────────────────────┘

Internal Design

• L0 / L1 / L2 Dynamic Folding — every memory stores 3 granularity layers (one-liner / bullet summary / full content); retrieval dynamically selects which layer to return based on relevance score and token budget

• Weibull Decay + Emotion Modulation — memories decay along a parametric Weibull curve; importance scores modulate the half-life, and emotional salience extends it further (up to 30%)

• Vector Pre-filter + LLM Dedup — 90% of dedup decisions use cheap cosine similarity (>= 0.92); only borderline cases invoke LLM judgment, keeping costs low without sacrificing accuracy

• Category-Aware Merge Strategies — profile and preferences use merge-on-conflict (latest wins); events and cases use append-only (history preserved)

• Display Score vs Elimination Score — dual-track retrieval: tier floor prevents core memories from ever dropping out, while decay boost lets fresh memories surface temporarily without permanently displacing stable ones

Full architecture deep-dive: docs/architecture.md

Interfaces

RecallNest serves two interfaces:

• MCP — for Claude Code, Gemini CLI, and Codex (native tool access)

• HTTP API — for custom agents, SDK-based apps, and any HTTP client

Agent framework examples

Examples live in integrations/examples/:

Base URL: http://localhost:4318

Full documentation: docs/api-reference.md

# Search & explore
bun run src/cli.ts search "your query"
bun run src/cli.ts explain "your query" --profile debug
bun run src/cli.ts distill "topic" --profile writing
bun run src/cli.ts stats

# Workflow observation
bun run src/cli.ts workflow-observe resume_context "Fresh window skipped continuity recovery." --outcome missed --scope project:recallnest
bun run src/cli.ts workflow-health resume_context --scope project:recallnest
bun run src/cli.ts workflow-evidence checkpoint_session --scope project:recallnest

# Conflict management
bun run src/cli.ts conflicts list
bun run src/cli.ts conflicts list --attention resolved
bun run src/cli.ts conflicts list --group-by cluster --attention resolved
bun run src/cli.ts conflicts audit
bun run src/cli.ts conflicts audit --export --format md
bun run src/cli.ts conflicts escalate --attention stale
bun run src/cli.ts conflicts show af70545a
bun run src/cli.ts conflicts resolve af70545a --keep-existing
bun run src/cli.ts conflicts resolve af70545a --merge
bun run src/cli.ts conflicts resolve --all --keep-existing --status open

# Memory health & visualization
bun run src/cli.ts lint                         # memory quality report
bun run src/cli.ts lint --scope project:myapp   # lint a specific scope
bun run src/cli.ts graph --open                 # export & open knowledge graph
bun run src/cli.ts graph --max-nodes 50         # smaller graph

# Ingestion & diagnostics
bun run src/cli.ts ingest --source all
bun run src/cli.ts doctor

Multilingual Support

RecallNest works out of the box with English. For multilingual memory (Chinese, Japanese, Thai, and 20+ more), install babel-memory with the language packs you need:

# Chinese
npm install babel-memory jieba-wasm

# Japanese
npm install babel-memory @sglkc/kuromoji

# Thai
npm install babel-memory wordcut

# European languages (German, French, Spanish, Russian, etc.)
npm install babel-memory snowball-stemmers

# Multiple languages at once
npm install babel-memory jieba-wasm @sglkc/kuromoji snowball-stemmers

RecallNest auto-detects babel-memory at startup — no configuration needed. Without babel-memory, RecallNest still works perfectly with standard BM25 text search.

Project Status & Roadmap

RecallNest is actively maintained. All major architecture phases are complete — see the full Roadmap for current priorities and future plans.

Relationship to memory-lancedb-pro

RecallNest started as a fork of memory-lancedb-pro and shares its core ideas around hybrid retrieval, decay modeling, and memory-as-engineering-system. The key difference:

• memory-lancedb-pro is an OpenClaw plugin — it adds long-term memory to a single OpenClaw agent.

• RecallNest is a standalone memory layer — it serves Claude Code, Codex, and Gemini CLI simultaneously through MCP + HTTP API, with session continuity, structured assets, and conflict management built in.

Credit

Special thanks to Qin Chao (@win4r) and the CortexReach team for the foundational work.

Part of the 小试AI open-source AI workflow:

License

MIT