TIMPs is an open-source AI memory agent that remembers what matters across sessions. It provides structured long-term memory with semantic retrieval, project isolation, and reflection-based storage.

TIMPs (Trustworthy Interactive Memory Partner System) is an open-source AI memory agent that revolutionizes how users interact with language models by providing persistent contextual memory across sessions. Designed for developers, researchers, and anyone who uses AI assistants repeatedly, it remembers facts, goals, preferences, and reflections from every conversation. The core value is eliminating the friction of reestablishing context each session: users describe their tech stack, project details, or personal preferences once, and TIMPs stores this information in a structured 14-field schema with importance scoring and vector embeddings. It can run fully locally with Ollama for zero cloud dependencies or integrate with OpenAI GPT-4 and Google Gemini. This flexibility makes it suitable for both privacy-critical and cloud-optimized environments, all while being free and open source under the MIT license.

The primary problem that TIMPs addresses is the lack of persistent context in standard AI chat interfaces. Users engaging in long-term projects often need to repeat themselves across sessions, which is inefficient and error-prone. TIMPs eliminates this friction by automatically recording and retrieving contextual information. After each conversation turn, the Reflection component analyzes the exchange, identifies key facts, and stores them with an importance score ranging from one to five stars. These memories are deposited into a PostgreSQL LongTermStore for exact keyword matching and a Qdrant VectorStore for semantic similarity search. When a user returns to a session, they simply use the `!blame` command to search for relevant memories. The dual retrieval shows both literal matches and semantically similar entries, along with their importance and retrieval frequency. This ensures that even if the exact phrasing changes, the system can surface the right context, making ongoing interactions seamless and highly productive.

A standout feature of TIMPs is its persistent memory system built on a 14-field schema that organizes each stored fact by type (explicit, reflection, goal, preference, etc.), importance score (one to five stars), project context, timestamp, and retrieval count. This rich metadata enables nuanced retrieval: for example, high-importance reflections are given priority over low-importance explicit facts. The Reflection component runs automatically after every conversation turn, analyzing the exchange to extract key information and assign an importance score. Users also have the option to explicitly store memories by stating them directly. All memories are stored simultaneously in PostgreSQL for exact SQL matching and in Qdrant for vector similarity, ensuring both precise and semantic recall. The `!audit` command provides a transparent view of the last ten stored memories with full metadata, allowing users to verify what has been learned and manage their memory store effectively.

TIMPs

Key Features

Use Cases

Who is this for?

Comments