vestige/README.md

Vestige

The cognitive engine that gives AI agents a brain.

Your Agent forgets everything between sessions. Vestige fixes that.

Built on 130 years of memory research — FSRS-6 spaced repetition, prediction error gating, synaptic tagging, spreading activation, memory dreaming — all running in a single Rust binary with a 3D neural visualization dashboard. 100% local. Zero cloud.

Quick Start | Dashboard | How It Works | Tools | Docs

---

What's New in v2.0.9 "Autopilot"

Autopilot flips Vestige from passive memory library to self-managing cognitive surface. Same 24 MCP tools, zero schema changes — but the moment you upgrade, 14 previously dormant cognitive primitives start firing on live events without any tool call from your client.

• One supervised backend task subscribes to the 20-event WebSocket bus and routes six event classes into the cognitive engine: MemoryCreated triggers synaptic-tagging PRP + predictive-access records, SearchPerformed warms the speculative-retrieval model, MemoryPromoted fires activation spread, MemorySuppressed emits the Rac1 cascade wave, high-importance ImportanceScored (>0.85) auto-promotes, and Heartbeat rate-limit-fires find_duplicates on large DBs. The engine mutex is never held across .await, so MCP dispatch is never starved.
• Panic-resilient supervisors. Both background tasks run inside an outer supervisor loop — if one handler panics on a bad memory, the supervisor respawns it in 5 s instead of losing every future event.
• Fully backward compatible. No new MCP tools. No schema migration. Existing v2.0.8 databases open without a single step. Opt out with VESTIGE_AUTOPILOT_ENABLED=0 if you want the passive-library contract back.
• 3,091 LOC of orphan v1.0 tool code removed — nine superseded modules (checkpoint, codebase, consolidate, ingest, intentions, knowledge, recall, plus helpers) verified zero non-test callers before deletion. Tool surface unchanged.

What's New in v2.0.8 "Pulse"

v2.0.8 wires the dashboard through to the cognitive engine. Eight new surfaces expose the reasoning stack visually — every one was MCP-only before.

• Reasoning Theater (/reasoning) — Cmd+K Ask palette over the 8-stage deep_reference pipeline (hybrid retrieval → cross-encoder rerank → spreading activation → FSRS-6 trust → temporal supersession → contradiction analysis → relation assessment → template reasoning chain). Evidence cards, confidence meter, contradiction geodesic arcs, superseded-memory lineage, evolution timeline. Zero LLM calls, 100% local.
• Pulse InsightToast — real-time toasts for DreamCompleted, ConsolidationCompleted, ConnectionDiscovered, promote/demote/suppress/unsuppress, Rac1CascadeSwept. Rate-limited, auto-dismiss, click-to-dismiss.
• Memory Birth Ritual (Terrarium) — new memories materialize in the 3D graph on every MemoryCreated: elastic scale-in, quadratic Bezier flight path, glow sprite fade-in, Newton's Cradle docking recoil. 60-frame sequence, zero-alloc math.
• 7 more dashboard surfaces — /duplicates, /dreams, /schedule, /importance, /activation, /contradictions, /patterns. Left nav expanded 8 → 16 with single-key shortcuts.
• Intel Mac (x86_64-apple-darwin) support restored via the ort-dynamic Cargo feature + Homebrew onnxruntime. Microsoft deprecated x86_64 macOS prebuilts; the dynamic-link path sidesteps that permanently. Closes #41.
• Contradiction-detection false positives eliminated — four thresholds tightened so adjacent-domain memories no longer flag as conflicts. On an FSRS-6 query this collapses false contradictions 12 → 0 without regressing legitimate test cases.

What's New in v2.0.7 "Visible"

Hygiene release closing two UI gaps and finishing schema cleanup. No breaking changes, no user-data migrations.

• POST /api/memories/{id}/suppress + /unsuppress HTTP endpoints — dashboard can trigger Anderson 2025 SIF + Rac1 cascade without dropping to raw MCP. suppressionCount, retrievalPenalty, reversibleUntil, labileWindowHours all in response. Suppress button joins Promote / Demote / Delete on the Memories page.
• Uptime in the sidebar footer — the Heartbeat event has carried uptime_secs since v2.0.5 but was never rendered. Now shows as up 3d 4h / up 18m / up 47s.
• execute_export panic fix — unreachable match arm replaced with a clean "unsupported export format" error instead of unwinding through the MCP dispatcher.
• predict surfaces predict_degraded: true on lock poisoning instead of silently returning empty vecs. memory_changelog honors start / end bounds. intention check honors include_snoozed.
• Migration V11 — drops dead knowledge_edges + compressed_memories tables (added speculatively in V4, never used).

What's New in v2.0.6 "Composer"

v2.0.6 is a polish release that makes the existing cognitive stack finally feel alive in the dashboard and stays out of your way on the prompt side.

• Six live graph reactions, not one — MemorySuppressed, MemoryUnsuppressed, Rac1CascadeSwept, Connected, ConsolidationStarted, and ImportanceScored now light the 3D graph in real time. v2.0.5 shipped suppress but the graph was silent when you called it; consolidation and importance scoring have been silent since v2.0.0. No longer.
• Intentions page actually works — fixes a long-standing bug where every intention rendered as "normal priority" (type/schema drift between backend and frontend) and context/time triggers surfaced as raw JSON.
• Opt-in composition mandate — the new MCP instructions string stays minimal by default. Opt in to the full Composing / Never-composed / Recommendation composition protocol with VESTIGE_SYSTEM_PROMPT_MODE=full when you want it, and nothing is imposed on your sessions when you don't.

What's New in v2.0.5 "Intentional Amnesia"

The first shipped AI memory system with top-down inhibitory control over retrieval. Other systems implement passive decay — memories fade if you don't touch them. Vestige v2.0.5 also implements active suppression: the new suppress tool compounds a retrieval penalty on every call (up to 80%), a background Rac1 worker fades co-activated neighbors over 72 hours, and the whole thing is reversible within a 24-hour labile window. Never deletes. The memory is inhibited, not erased.

Ebbinghaus 1885 models what happens to memories you don't touch. Anderson 2025 models what happens when you actively want to stop thinking about one. Every other AI memory system implements the first. Vestige is the first to ship the second.

Based on Anderson et al. 2025 (Suppression-Induced Forgetting, Nat Rev Neurosci) and Cervantes-Sandoval et al. 2020 (Rac1 synaptic cascade). 24 tools · 30 cognitive modules · 1,223 tests.

Earlier releases (v2.0 "Cognitive Leap" → v2.0.4 "Deep Reference")

• v2.0.4 — deep_reference Tool — 8-stage cognitive reasoning pipeline with FSRS-6 trust scoring, intent classification, spreading activation, contradiction analysis, and pre-built reasoning chains. Token budgets raised 10K → 100K. CORS tightened.
• v2.0 — 3D Memory Dashboard — SvelteKit + Three.js neural visualization with real-time WebSocket events, bloom post-processing, force-directed graph layout.
• v2.0 — WebSocket Event Bus — Every cognitive operation broadcasts events: memory creation, search, dreaming, consolidation, retention decay.
• v2.0 — HyDE Query Expansion — Template-based Hypothetical Document Embeddings for dramatically improved search quality on conceptual queries.
• v2.0 — Nomic v2 MoE (experimental) — fastembed 5.11 with optional Nomic Embed Text v2 MoE (475M params, 8 experts) + Metal GPU acceleration.
• v2.0 — Command Palette — Cmd+K navigation, keyboard shortcuts, responsive mobile layout, PWA installable.
• v2.0 — FSRS Decay Visualization — SVG retention curves with predicted decay at 1d/7d/30d.

---

Quick Start

# 1. Install (macOS Apple Silicon)
curl -L https://github.com/samvallad33/vestige/releases/latest/download/vestige-mcp-aarch64-apple-darwin.tar.gz | tar -xz
sudo mv vestige-mcp vestige vestige-restore /usr/local/bin/

# 2. Connect to Claude Code
claude mcp add vestige vestige-mcp -s user

# Or connect to Codex
codex mcp add vestige -- /usr/local/bin/vestige-mcp

# 3. Test it
# "Remember that I prefer TypeScript over JavaScript"
# ...new session...
# "What are my coding preferences?"
# → "You prefer TypeScript over JavaScript."

Other platforms & install methods

Linux (x86_64):

curl -L https://github.com/samvallad33/vestige/releases/latest/download/vestige-mcp-x86_64-unknown-linux-gnu.tar.gz | tar -xz
sudo mv vestige-mcp vestige vestige-restore /usr/local/bin/

macOS (Intel): Microsoft is discontinuing x86_64 macOS prebuilts after ONNX Runtime v1.23.0, so Vestige's Intel Mac build links dynamically against a Homebrew-installed ONNX Runtime via the ort-dynamic feature. Install with:

brew install onnxruntime
curl -L https://github.com/samvallad33/vestige/releases/latest/download/vestige-mcp-x86_64-apple-darwin.tar.gz | tar -xz
sudo mv vestige-mcp vestige vestige-restore /usr/local/bin/
echo 'export ORT_DYLIB_PATH="'"$(brew --prefix onnxruntime)"'/lib/libonnxruntime.dylib"' >> ~/.zshrc
source ~/.zshrc
claude mcp add vestige vestige-mcp -s user

Full Intel Mac guide (build-from-source + troubleshooting): docs/INSTALL-INTEL-MAC.md.

Windows: Prebuilt binaries ship but usearch 2.24.0 hit an MSVC compile break (usearch#746); we've pinned =2.23.0 until upstream fixes it. Source builds work with:

git clone https://github.com/samvallad33/vestige && cd vestige
cargo build --release -p vestige-mcp

npm:

npm install -g vestige-mcp-server

Build from source (requires Rust 1.91+):

git clone https://github.com/samvallad33/vestige && cd vestige
cargo build --release -p vestige-mcp
# Optional: enable Metal GPU acceleration on Apple Silicon
cargo build --release -p vestige-mcp --features metal

---

Works Everywhere

Vestige speaks MCP — the universal protocol for AI tools. One brain, every IDE.

IDE	Setup
Claude Code	claude mcp add vestige vestige-mcp -s user
Codex	Integration guide
Claude Desktop	2-min setup
Xcode 26.3	Integration guide
Cursor	Integration guide
VS Code (Copilot)	Integration guide
JetBrains	Integration guide
Windsurf	Integration guide

---

🧠 3D Memory Dashboard

Vestige v2.0 ships with a real-time 3D visualization of your AI's memory. Every memory is a glowing node in 3D space. Watch connections form, memories pulse when accessed, and the entire graph come alive during dream consolidation.

Features:

• Force-directed 3D graph with 1000+ nodes at 60fps
• Bloom post-processing for cinematic neural network aesthetic
• Real-time WebSocket events: memories pulse on access, burst on creation, fade on decay
• Dream visualization: graph enters purple dream mode, replayed memories light up sequentially
• FSRS retention curves: see predicted memory decay at 1d, 7d, 30d
• Command palette (Cmd+K), keyboard shortcuts, responsive mobile layout
• Installable as PWA for quick access

Tech: SvelteKit 2 + Svelte 5 + Three.js + Tailwind CSS 4 + WebSocket

The dashboard runs automatically at http://localhost:3927/dashboard when the MCP server starts.

---

Architecture

┌─────────────────────────────────────────────────────┐
│  SvelteKit Dashboard (apps/dashboard)                │
│  Three.js 3D Graph · WebGL + Bloom · Real-time WS   │
├─────────────────────────────────────────────────────┤
│  Axum HTTP + WebSocket Server (port 3927)            │
│  15 REST endpoints · WS event broadcast              │
├─────────────────────────────────────────────────────┤
│  MCP Server (stdio JSON-RPC)                         │
│  24 tools · 30 cognitive modules                     │
├─────────────────────────────────────────────────────┤
│  Cognitive Engine                                    │
│  ┌──────────┐ ┌──────────┐ ┌───────────────┐       │
│  │ FSRS-6   │ │ Spreading│ │ Prediction    │       │
│  │ Scheduler│ │ Activation│ │ Error Gating  │       │
│  └──────────┘ └──────────┘ └───────────────┘       │
│  ┌──────────┐ ┌──────────┐ ┌───────────────┐       │
│  │ Memory   │ │ Synaptic │ │ Hippocampal   │       │
│  │ Dreamer  │ │ Tagging  │ │ Index         │       │
│  └──────────┘ └──────────┘ └───────────────┘       │
├─────────────────────────────────────────────────────┤
│  Storage Layer                                       │
│  SQLite + FTS5 · USearch HNSW · Nomic Embed v1.5    │
│  Optional: Nomic v2 MoE · Qwen3 Reranker · Metal   │
└─────────────────────────────────────────────────────┘

---

Why Not Just Use RAG?

RAG is a dumb bucket. Vestige is an active organ.

	RAG / Vector Store	Vestige
Storage	Store everything	Prediction Error Gating — only stores what's surprising or new
Retrieval	Nearest-neighbor	7-stage pipeline — HyDE expansion + reranking + spreading activation
Decay	Nothing expires	FSRS-6 — memories fade naturally, context stays lean
Forgetting (v2.0.5)	Delete only	suppress tool — compounding top-down inhibition, neighbor cascade, reversible 24h
Duplicates	Manual dedup	Self-healing — auto-merges "likes dark mode" + "prefers dark themes"
Importance	All equal	4-channel scoring — novelty, arousal, reward, attention
Sleep	No consolidation	Memory dreaming — replays, connects, synthesizes insights
Health	No visibility	Retention dashboard — distributions, trends, recommendations
Visualization	None	3D neural graph — real-time WebSocket-powered Three.js
Privacy	Usually cloud	100% local — your data never leaves your machine

---

🔬 The Cognitive Science Stack

This isn't a key-value store with an embedding model bolted on. Vestige implements real neuroscience:

Prediction Error Gating — The hippocampal bouncer. When new information arrives, Vestige compares it against existing memories. Redundant? Merged. Contradictory? Superseded. Novel? Stored with high synaptic tag priority.

FSRS-6 Spaced Repetition — 21 parameters governing the mathematics of forgetting. Frequently-used memories stay strong. Unused memories naturally decay. Your context window stays clean.

HyDE Query Expansion (v2.0) — Template-based Hypothetical Document Embeddings. Expands queries into 3-5 semantic variants, embeds all variants, and searches with the centroid embedding for dramatically better recall on conceptual queries.

Synaptic Tagging — A memory that seemed trivial this morning can be retroactively tagged as critical tonight. Based on Frey & Morris, 1997.

Spreading Activation — Search for "auth bug" and find the related JWT library update from last week. Memories form a graph, not a flat list. Based on Collins & Loftus, 1975.

Dual-Strength Model — Every memory has storage strength (encoding quality) and retrieval strength (accessibility). A deeply stored memory can be temporarily hard to retrieve — just like real forgetting. Based on Bjork & Bjork, 1992.

Memory Dreaming — Like sleep consolidation. Replays recent memories to discover hidden connections, strengthen important patterns, and synthesize insights. Dream-discovered connections persist to a graph database. Based on the Active Dreaming Memory framework.

Waking SWR Tagging — Promoted memories get sharp-wave ripple tags for preferential replay during dream consolidation. 70/30 tagged-to-random ratio. Based on Buzsaki, 2015.

Autonomic Regulation — Self-regulating memory health. Auto-promotes frequently accessed memories. Auto-GCs low-retention memories. Consolidation triggers on 6h staleness or 2h active use.

Active Forgetting (v2.0.5) — Top-down inhibitory control via the suppress tool. Other memory systems implement passive decay — the Ebbinghaus 1885 "use it or lose it" curve, sometimes with trust-weighted strength factors. Vestige v2.0.5 also implements active top-down suppression: each suppress call compounds (Suppression-Induced Forgetting, Anderson 2025), a background Rac1 cascade worker fades co-activated neighbors across the connection graph (Cervantes-Sandoval & Davis 2020), and a 24-hour labile window allows reversal (Nader reconsolidation semantics on a pragmatic axis). The memory persists — it's inhibited, not erased. Explicitly distinct from Anderson 1994 retrieval-induced forgetting (bottom-up, passive competition during retrieval), which is a separate, older primitive that several other memory systems implement. Based on Anderson et al., 2025 and Cervantes-Sandoval et al., 2020. First shipped AI memory system with this primitive.

Full science documentation ->

---

🛠 24 MCP Tools

Context Packets

Tool	What It Does
session_context	One-call session init — replaces 5 calls with token-budgeted context, automation triggers, expandable IDs

Core Memory

Tool	What It Does
search	7-stage cognitive search — HyDE expansion + keyword + semantic + reranking + temporal + competition + spreading activation
smart_ingest	Intelligent storage with CREATE/UPDATE/SUPERSEDE via Prediction Error Gating. Batch mode for session-end saves
memory	Get, delete, check state, promote (thumbs up), demote (thumbs down)
codebase	Remember code patterns and architectural decisions per-project
intention	Prospective memory — "remind me to X when Y happens"

Cognitive Engine

Tool	What It Does
dream	Memory consolidation — replays memories, discovers connections, synthesizes insights, persists graph
explore_connections	Graph traversal — reasoning chains, associations, bridges between memories
predict	Proactive retrieval — predicts what you'll need next based on context and activity

Autonomic

Tool	What It Does
memory_health	Retention dashboard — distribution, trends, recommendations
memory_graph	Knowledge graph export — force-directed layout, up to 200 nodes

Scoring & Dedup

Tool	What It Does
importance_score	4-channel neuroscience scoring (novelty, arousal, reward, attention)
find_duplicates	Detect and merge redundant memories via cosine similarity

Maintenance

Tool	What It Does
system_status	Combined health + stats + cognitive state + recommendations
consolidate	Run FSRS-6 decay cycle (also auto-runs every 6 hours)
memory_timeline	Browse chronologically, grouped by day
memory_changelog	Audit trail of state transitions
backup / export / gc	Database backup, JSON export, garbage collection
restore	Restore from JSON backup

Deep Reference (v2.0.4)

Tool	What It Does
deep_reference	Cognitive reasoning across memories. 8-stage pipeline: FSRS-6 trust scoring, intent classification, spreading activation, temporal supersession, contradiction analysis, relation assessment, dream insight integration, and algorithmic reasoning chain generation. Returns trust-scored evidence with a pre-built reasoning scaffold.
cross_reference	Backward-compatible alias for deep_reference.

Active Forgetting (v2.0.5)

Tool

What It Does

suppress

Top-down active forgetting — neuroscience-grounded inhibitory control over retrieval. Distinct from memory.delete (destroys the row) and memory.demote (one-shot ranking hit). Each call compounds a retrieval-score penalty (Anderson 2025 SIF), and a background Rac1 cascade worker fades co-activated neighbors over 72h (Davis 2020). Reversible within a 24-hour labile window via reverse: true. The memory persists — it is inhibited, not erased.

---

Make Your AI Use Vestige Automatically

Add this to your CLAUDE.md:

## Memory

At the start of every session:
1. Search Vestige for user preferences and project context
2. Save bug fixes, decisions, and patterns without being asked
3. Create reminders when the user mentions deadlines

You Say	AI Does
"Remember this"	Saves immediately
"I prefer..." / "I always..."	Saves as preference
"Remind me..."	Creates a future trigger
"This is important"	Saves + promotes

Full CLAUDE.md templates ->

---

Technical Details

Metric	Value
Language	Rust 2024 edition (MSRV 1.91)
Codebase	80,000+ lines, 1,292 tests (366 core + 425 mcp + 497 e2e + 4 doctests)
Binary size	~20MB
Embeddings	Nomic Embed Text v1.5 (768d → 256d Matryoshka, 8192 context)
Vector search	USearch HNSW (20x faster than FAISS)
Reranker	Jina Reranker v1 Turbo (38M params, +15-20% precision)
Storage	SQLite + FTS5 (optional SQLCipher encryption)
Dashboard	SvelteKit 2 + Svelte 5 + Three.js + Tailwind CSS 4
Transport	MCP stdio (JSON-RPC 2.0) + WebSocket
Cognitive modules	30 stateful (17 neuroscience, 11 advanced, 2 search)
First run	Downloads embedding model (~130MB), then fully offline
Platforms	macOS ARM + Intel + Linux x86_64 + Windows x86_64 (all prebuilt). Intel Mac needs brew install onnxruntime — see install guide.

Optional Features

# Metal GPU acceleration (Apple Silicon — faster embedding inference)
cargo build --release -p vestige-mcp --features metal

# Nomic Embed Text v2 MoE (475M params, 305M active, 8 experts)
cargo build --release -p vestige-mcp --features nomic-v2

# Qwen3 Reranker (Candle backend, high-precision cross-encoder)
cargo build --release -p vestige-mcp --features qwen3-reranker

# SQLCipher encryption
cargo build --release -p vestige-mcp --no-default-features --features encryption,embeddings,vector-search

---

CLI

vestige stats                    # Memory statistics
vestige stats --tagging          # Retention distribution
vestige stats --states           # Cognitive state breakdown
vestige health                   # System health check
vestige consolidate              # Run memory maintenance
vestige restore <file>           # Restore from backup
vestige dashboard                # Open 3D dashboard in browser

---

Documentation

Document	Contents
FAQ	30+ common questions answered
Science	The neuroscience behind every feature
Storage Modes	Global, per-project, multi-instance
CLAUDE.md Setup	Templates for proactive memory
Configuration	CLI commands, environment variables
Integrations	Codex, Xcode, Cursor, VS Code, JetBrains, Windsurf
Changelog	Version history

---

Troubleshooting

"Command not found" after installation

Ensure vestige-mcp is in your PATH:

which vestige-mcp
# Or use the full path:
claude mcp add vestige /usr/local/bin/vestige-mcp -s user

Embedding model download fails

First run downloads ~130MB from Hugging Face. If behind a proxy:

export HTTPS_PROXY=your-proxy:port

Cache: macOS ~/Library/Caches/com.vestige.core/fastembed | Linux ~/.cache/vestige/fastembed

Dashboard not loading

The dashboard starts automatically on port 3927 when the MCP server runs. Check:

curl http://localhost:3927/api/health
# Should return {"status":"healthy",...}

More troubleshooting ->

---

Contributing

Issues and PRs welcome. See CONTRIBUTING.md.

License

AGPL-3.0 — free to use, modify, and self-host. If you offer Vestige as a network service, you must open-source your modifications.

---

Built by @samvallad33
_{80,000+ lines of Rust · 30 cognitive modules · 130 years of memory research · one 22MB binary}