Vestige
A cognitive-memory system for AI agents, distributed as a single ~22MB Rust binary that doubles as an **MCP server** for Claude, Cursor, VS Code, Xcode, and JetBrains. Repository: [github.com/samvallad33/vestige](https://github.com/samvallad33/vestige). Vestige's core is the **FSRS-6 spaced-repetition algorithm** (Free Spaced Repetition Scheduler), with 29 distinct "brain modules" mapping to different cognitive operations (novelty, arousal, reward, attention) — turning memory into an actively-managed cognitive resource rather than passive storage.
Definition
A cognitive-memory system for AI agents, distributed as a single ~22MB Rust binary that doubles as an **MCP server** for Claude, Cursor, VS Code, Xcode, and JetBrains. Repository: [github.com/samvallad33/vestige](https://github.com/samvallad33/vestige). Vestige's core is the **FSRS-6 spaced-repetition algorithm** (Free Spaced Repetition Scheduler), with 29 distinct "brain modules" mapping to different cognitive operations (novelty, arousal, reward, attention) — turning memory into an actively-managed cognitive resource rather than passive storage.
Standard memory systems retrieve based on semantic similarity, ignoring how human cognition actually weights memories (recency, emotional salience, frequency of use, distinctiveness). Vestige imports cognitive-science scheduling primitives (FSRS-6, originally an Anki algorithm) into AI memory, so the system surfaces information based on a multi-channel scoring function rather than just vector distance. The MCP-server delivery means it integrates into any agentic IDE or platform that speaks MCP.
Cognitive-memory layer for IDE-based AI agents, FSRS-6-managed long-term context for coding assistants, multi-channel memory scoring (novelty + arousal + reward + attention), MCP-server-delivered memory for the Anthropic / Cursor / VS Code ecosystem.
Recent developments
- "Cognitive Sandwich" architecture: pre-think + post-speak gating. Vestige now ships memory injection before the model thinks plus optional Sanhedrin veto after the model speaks — bracketing model output with cognitive primitives on both sides. Per GitHub — samvallad33/vestige README.
- All Claude Code hooks opt-in by default as of May 1, 2026. Sanhedrin, preflight, and Claude Code hook integrations all default off; users opt in with
--enable-preflight/--enable-sanhedrinflags. Safer default + power-user upgrade path. Per Vestige CHANGELOG. - v2.1.1: cross-machine memory migration + opt-in Qwen3 embeddings. Moves memories between machines without losing cognitive state; opt-in to Qwen3 local-embedding model for higher-recall retrieval (vs the default smaller embedder). Per Vestige releases.
- 130 years of memory research baked in. Vestige-2 fork advertises spreading-activation, synaptic-tagging, and prediction-error-gating mechanisms from cognitive neuroscience — these are the "29 brain modules" framing operationalized as memory primitives. Per GitHub — nerdyviking321/vestige-2.
- 100% local memory server — privacy positioning. Vestige's product framing leans on the "memory never leaves your machine" guarantee — distinct value-prop vs Mem0/Zep managed-cloud offerings. Targets the privacy-first dev/coding-assistant slice. Per MCPMarket — Vestige: Advanced Persistent Memory for Claude AI.
- Multi-IDE MCP integration: Claude Desktop, Cursor, VS Code, Xcode, JetBrains. Ships docs + setup guides for each — the IDE-MCP cohort is Vestige's primary distribution surface, not a standalone web app. Per Vestige — VS Code integration docs and Vestige — JetBrains integration docs.
Connections 8
Outbound 7
scoped_to2implements1acts_as1integrates_with1Inbound 1
used_by1