TIMPs (Trustworthy Interactive Memory Partner System) is an open-source AI agent with persistent memory. It stores facts, preferences, goals, and reflections from your conversations in PostgreSQL and Qdrant, then retrieves relevant context automatically in future sessions.

The product features persistent memory with a 14-field schema storing facts, goals, preferences, and reflections with full audit trail. It offers dual-mode search combining SQL keyword matching (ILIKE) and Qdrant vector similarity for precise retrieval. Memories are scoped per project with project isolation preventing cross-contamination between codebases or clients. Ephemeral mode allows private sessions that leave zero persistent trace. TIMPs is model agnostic, supporting OpenAI GPT-4, Google Gemini, or local Ollama with provider switching via `--provider` flag. It provides multiple interfaces including REST API, TUI, and Web interfaces.

After each turn, TIMPs reflects on the conversation and extracts facts with an importance score (1–5). Facts are stored with 14 metadata fields in PostgreSQL and as vector embeddings in Qdrant. Future queries trigger dual-mode retrieval: SQL keyword matching + cosine vector similarity. The architecture includes a Core Agent with system prompt and tool definitions, Planner to break goals into steps, Executor to run tools (file, web, search), and Reflection to extract and score memories post-turn. Memory System consists of ShortTermStore with rolling context window (token budget), LongTermStore with PostgreSQL (14-field schema, ILIKE search), and VectorStore with Qdrant cosine similarity (1536-dim embeddings).

TIMPs remembers what matters, retrieves context when relevant, and evolves with every conversation. It enables AI agents that actually know you across every session, maintaining context across interactions.

The product is built for developers who care about their tools actually remembering context. It supports integration through REST API, TUI, and Web interfaces, with technical implementation using TypeScript, PostgreSQL, and Qdrant. Users can run it fully locally with Ollama for zero cloud calls, with all data living in their own PostgreSQL and Qdrant instances.