Vestige is an open-source agent harness and persistent memory infrastructure framework developed by samvallad33. Designed specifically for AI coding agents and autonomous workflows, Vestige focuses on structured agent memory management, temporal state tracking, and memory evaluation benchmarks. Vestige operates as an agent-neutral memory layer and evaluation suite. It interfaces with autonomous agents across command-line environments and provides evaluation tooling, embedding profiles, and an observatory dashboard built with Svelte for monitoring agent recall, memory access, and state transitions over time. What distinguishes Vestige from standard coding agent harnesses is its explicit focus on rigorous memory fidelity. Instead of relying purely on naive vector retrieval or monolithic context windows, it provides structured benchmarking harnesses (such as agent-memory-eval), memory access auditability (including protections against stale state resurrection), and reversible embedding profiles to prevent data corruption across agent sessions. Vestige is distributed as an open-source project on GitHub under active development, suitable for developers building, testing, and benchmarking long-running autonomous coding agents.
Agent-neutral memory harness — Decouples memory orchestration and retrieval from specific agent runtimes. - Agent memory evaluation framework — Built-in benchmarking tools (agent-memory-eval) to test and measure agent recall accuracy. - Reversible embedding profiles — Explicit embedding configurations designed for consistent retrieval and state migration. - Observatory web dashboard — Svelte-based monitoring interface to visualize memory scenes, retrieval traces, and verdict cards. - Temporal state management — Prevents stale memory resurrection and accurately tracks access history across session lifecycles. - Blackbox audit safety — Built-in security gating against destructive writes and automated PR content redaction.
Use cases
Use Case: Benchmarking Agent Long-Term Memory Developers evaluating multi-session coding agents can run Vestige's benchmark suites to measure precision and recall over time, validating whether agents maintain context across large codebases without hallucinating prior states.
Use Case: Memory Layer for Custom Agent Harnesses Teams building custom terminal or background coding agents can integrate Vestige as their underlying memory subsystem, utilizing its reversible embedding profiles and memory lifecycle management.
Use Case: Observability and Memory Tracing Engineers debugging agent failures can run the Vestige Observatory dashboard to inspect memory scenes, observe query evaluations, and identify why an agent retrieved or ignored specific contextual memories.
Who it is for
Vestige is aimed at AI systems engineers, agent harness developers, and researchers benchmarking long-term memory and retrieval performance in autonomous coding agents. It is not intended for end users looking for a simple plug-and-play CLI chatbot with zero configuration.
Install guide
Installation Clone the repository and install dependencies using Node.js/npm or pnpm: bash git clone https://github.com/samvallad33/vestige.git cd vestige npm install
First Run Configure your environment variables and embedding model configurations in .env, then start the development environment or dashboard: bash cp .env.example .env npm run dev
Example Session Run the agent memory evaluation suite to benchmark memory retrieval accuracy: bash npm run test:memory-eval
Beads — Beads (bd) is an open-source task coordination and state orchestration harness for AI coding agents developed by Gastown Hall. It…
Letta Code — Letta Code is an open-source terminal-based AI coding assistant developed by Letta (formerly MemGPT). Built on top of the Letta…
GoodMemory — GoodMemory is an open-source agent memory orchestration harness and context management system developed by hjqcan. Designed to give AI agents…
pi-mem — pi-mem is an open-source memory harness and persistent context manager developed by Jo Inc for the Pi CLI coding agent.…
SWE-agent — SWE-agent is an open-source autonomous software engineering agent harness built and maintained by researchers from Princeton University and Stanford University.…
OpenHands — OpenHands (formerly OpenDevin) is an open-source autonomous AI software development agent and orchestration platform. It provides an extensible runtime environment…