TIMPs — The open source AI memory agent

timps — interactive memory session

$timps cli --user-id 1 --provider openai

✓ PostgreSQL connected

✓ Qdrant vector store ready (1536-dim)

────────────────────────────────────────

you > I'm working on a React app with TypeScript and Zustand

timps › Got it! I'll remember your stack. React + TypeScript + Zustand sounds solid.

[REFLECTION stored · importance: ⭐⭐⭐⭐⭐ · id: #42]

you > !blame React

🔍 Found 2 memory item(s) matching "React":

[42] REFLECTION ⭐⭐⭐⭐⭐ — React + TypeScript + Zustand

[38] EXPLICIT ⭐⭐⭐ — prefers functional components

you >

What is TIMPs?

An AI agent that actually
knows you — across
every session.

[*] Persistent memory — 14-field schema stores facts, goals, preferences, and reflections with full audit trail.
[*] Dual-mode search — SQL keyword matching (ILIKE) + Qdrant vector similarity for precise retrieval.
[*] Project isolation — memories are scoped per project. No cross-contamination between codebases or clients.
[*] Ephemeral mode — toggle a private session that leaves zero persistent trace.
[*] Model agnostic — OpenAI GPT-4, Google Gemini, or local Ollama. Swap anytime via --provider.
[*] REST API + TUI + Web — use from terminal, desktop, or browser.

The open source AI memory agent

Built in the open, used by developers who care about their tools actually remembering context.

Fig 1.

Memory fields per stored fact

Fig 2.

Model providers supported (OpenAI, Gemini, Ollama)

Fig 3.

2×

Dual search — SQL + vector semantic retrieval

Architecture

User → TUI / CLI / Web / REST API
│
▼
Core Agent ← system prompt + tool definitions
│
├── Planner break goal into steps
├── Executor run tools (file, web, search)
└── Reflection extract + score memories post-turn
│
▼
Memory System
├── ShortTermStore rolling context window (token budget)
├── LongTermStore PostgreSQL (14-field schema, ILIKE search)
└── VectorStore Qdrant cosine similarity (1536-dim embeddings)

Memory Commands

!blame

Search all stored memories for a keyword. Runs dual SQL + vector search and shows importance scores and retrieval counts.

!forget

Preview matching memories before deletion. Requires confirmation. Atomically removes from PostgreSQL and Qdrant simultaneously.

!audit

View last 10 stored memories with full metadata — type, importance stars, created timestamp, retrieval count.

Built for privacy first

Your context
stays on your
machine.

Run fully local with Ollama — zero cloud calls
Ephemeral mode leaves no persistent storage
Project isolation prevents memory leakage between workspaces
All data lives in your own PostgreSQL and Qdrant instances
Open source — audit every line of the codebase

FAQ

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.

No. TIMPs works fully locally using Ollama. If you want to use GPT-4 or Gemini, add your key to the .env file. Switch providers anytime with the --provider flag.

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.

Launch with --mode ephemeral and nothing is persisted. The session exists only in RAM. Use it for sensitive conversations, debugging, or one-off queries you don't want stored.

Yes. TIMPs is released under the MIT license. Every line of the agent, memory system, and API is available on GitHub.

Yes. The web chat interface connects to your local TIMPs server running on localhost:3000. Run npm run server, then open the web UI. All commands work in the browser too.

The open sourceAI memory agent

The open source
AI memory agent