What problem does it solve? AI agents repeatedly re-read source code and documentation across chat sessions, inflating token consumption and losing architectural decisions after context compaction. Cogni persists compact semantic memory signatures in a local SQLite database so agents can recall prior decisions, bugfixes, and patterns without re-analyzing the codebase. ## Core Features & Use Cases - Two-Step Retrieval: Run a lightweight cogni search to find relevant memory IDs, then hydrate full content only for matching entries with cogni get, keeping context usage minimal. - High-Density Signatures: Save structured memories with What/Why/Where/Learned fields, deterministic topic keys, and automatic upserts to prevent duplicates. - Session Lifecycle Management: Persist end-of-session summaries and instantly restore context after compaction using cogni_session_summary and cogni_context. - Use Case: Before fixing a JWT authentication bug, an agent searches Cogni for prior resolutions in the auth module, retrieves the exact signature describing the previous fix, and applies the established pattern without re-reading hundreds of lines of code. ## Quick Start Ask your AI agent to search Cogni memory for previous decisions about the module you are working on before starting the task.
