AI coding agents are getting stronger, but the project memory around them still disappears when a chat gets long, a session ends, or you switch tools. I built Rta-Smriti Brain to move that memory out of the chat and into the project. It combines repository structure, durable decisions, long-session handoffs, and task-specific context packs in a local SQLite brain. The operator console lets you inspect the graph, browse indexed files, review memories, check freshness, and prepare context for any agent. The unusual part is the Pramana evidence model. An observed test result, a trusted human instruction, an inference, a prior memory, and a hypothesis remain different kinds of knowledge instead of becoming one undifferentiated memory blob. This is an alpha developer release. It is open source, local-first, and already tested across six project brains, including a 26,482-file repository. I would especially value feedback on onboarding, retrieval quality, and integrations with your preferred coding agent.

Rta-Smriti Brain

Stop re-explaining your project to every new AI chat.

Stop re-explaining your project to every new AI chat.