AllSource gives AI agents durable, event-sourced memory with complete provenance and time-travel queries. It recalls in ~12μs p99, ingests 469K events/sec, and exposes 73 MCP tools—without an external embedding API. Run the Apache-2.0 core yourself or use AllSource Prime for hosted, derived memory. Built for teams that need to reconstruct exactly what an agent knew, when it knew it, and which source shaped each decision. Self-host free; hosted Prime is £18.99/month after a 14-day trial.

Hey Product Hunt 👋 I built AllSource after watching AI agents fail in ways ordinary chat-history storage could not explain. Summaries drifted, provenance disappeared, and teams could not reconstruct what an agent knew before a bad decision. AllSource treats memory as an append-only event log. Every memory keeps its source and timestamp, time-travel queries reconstruct prior state, and the Rust engine recalls at ~12μs p99 while ingesting 469K events/sec. It exposes 73 MCP tools and needs no external embedding API. Core is Apache-2.0 and self-hostable. AllSource Prime adds hosted, derived memory for teams that want managed infrastructure. I’d love feedback from anyone building production agents: where does your memory layer fail today—durability, provenance, retrieval quality, or cost?

AllSource

Durable, time-travel memory for production AI agents

Durable, time-travel memory for production AI agents