Atlas
Local-first memory and evidence for AI agents
Start new thread

Atlas - Local-first memory and evidence for AI agents

by•
Atlas is local-first memory, state, knowledge and evidence infrastructure for AI agents. It keeps project knowledge persistent across agents, processes and sessions, with hybrid search, graph relationships, evidence, work state and an audit trail. Instead of rebuilding memory for every agent project, Atlas provides it as reusable infrastructure through MCP, REST and gRPC.

Add a comment

Replies

Best
Maker
📌
Hey Product Hunt 👋 I built Atlas as part of a broader idea behind Titans: foundational capabilities that AI agents repeatedly need shouldn't have to be redesigned and rebuilt inside every new project. They should exist as reusable infrastructure that agents and applications can simply consume. Atlas handles one of those recurring problems: persistent memory, state, knowledge and evidence. Agent sessions are temporary, but the projects they work on are not. Decisions, relationships, evidence, project state and accumulated knowledge often need to survive across sessions, processes and even different agents. Atlas provides a persistent local layer for that information, including: • Persistent project memory and state • Hybrid semantic and lexical search • Knowledge graph relationships • Evidence and provenance • Work state • Structured data and blobs • Audit history • MCP over stdio • Local REST and gRPC interfaces Atlas is local-first and agent-first, currently runs on Windows and Linux, has no telemetry, and is distributed through signed releases. It is one of the first systems in the wider Titans ecosystem. Its companion system, Cronus, provides durable execution: background jobs, workflows, scheduling, retries, checkpoints and recovery. Atlas remembers. Cronus runs. Both are designed to work independently or together. Existing applications, new agent systems and other parts of Titans can consume these capabilities instead of rebuilding them from scratch. Other tools are already being implemented as part of the ecosystem around the same principle: identify recurring infrastructure problems in agentic systems, solve them once as reusable capabilities, and make them available to future projects. I'd especially value feedback from people building agent systems: • How are you currently maintaining knowledge and project state across agent sessions? • Do multiple agents in your systems share the same persistent context? • What would you need from a memory and evidence layer before trusting it with real agent workflows? Critical technical feedback is very welcome.