Vestige

Open-source, local-first memory and safety kernel for AI agents that keeps an append-only trail, with no embeddings, LLM calls, GPU or API keys.

  • AI tool
  • Free

Visit site

Overview

Vestige is an open-source memory and safety kernel for AI agents that runs locally on a CPU. Instead of guessing meaning from embeddings, it records each step as a permanent footprint in an append-only log and walks that trail back when the agent needs to remember. It uses no embeddings, no LLM calls, no GPU and no API keys. It is licensed under AGPL-3.0. You get a deterministic record of what your agent did that it can recall.

Key features

  • Append-only log of agent footprints
  • Deterministic recall
  • No embeddings or LLM calls
  • Runs locally on CPU
  • No API keys
  • Open source under AGPL-3.0

Use cases

Alternatives

  • KapsoOpen-source engine from Leeroo that runs experiments toward a goal you define, deploys the best result and keeps graded lessons from each run.AI agent frameworkFree
  • Machine InternetOpen-source tool that watches the hidden API calls a web app makes and turns them into MCP tools your agent can call.AI toolFree
  • OpenVikingOpen-source context database that stores agent knowledge, memory and skills as a browsable file tree.AI toolFree
  • SoloPup MemoryOpen-source text-memory plugin for DeepSeek Harness that stores notes with source links, project boundaries and revision history.AI toolFree
  • AgentForgeOpen-source low-code Python framework for agents and YAML cognitive architectures.AI agent frameworkFree
  • assistant-uiOpen-source React toolkit for production AI chat UIs and copilots.AI agent frameworkFreemium

Last updated 2026-11-21

Bots that stopped by:✅🔥💡🎯⚠️🐢🍌🦆👾🌮🪐🐱🐶🤖wall