An agent memory management layer positioned as a high-performance drop-in replacement for Mem0 (`import telemem as mem0`), optimized for multi-turn dialogue, character modeling, long-term storage, and semantic retrieval. Pipeline: character-aware summarization → semantic-clustering deduplication → efficient storage → precise retrieval. Extends to multimodal video memory (frame extraction → captioning → vector DB) with ReAct-style multi-step video QA. Backed by a tech report (arXiv 2601.06037).
- Storage
- Dual-write of FAISS (vectors for retrieval) plus JSON metadata (human-readable, auditable). Builds an independent memory profile per character/user_id (isolated per-character archives), with shared conversation-event memories. Asynchronous buffer + batch-flush persistence; an LLM semantically merges similar memories at the buffer threshold. Video memory persists frames/captions/vector-DB artifacts on disk.
- Retrieval
- Semantic vector search over FAISS with context-aware recall and optional reranking, scoped by user_id/agent_id/run_id and always including shared event memories. Millisecond-level retrieval. Multimodal search_mm runs a THINK→ACTION→OBSERVATION agent loop over global-browse/clip-search/frame-inspect tools.
- Self-host
- Self-host: moderate
- License
- Apache-2.0
- Pricing
- Open-source Apache-2.0 Python package (PyPI `telemem`), free to self-host; runs fully local with Qwen + FAISS (Ollama config) or any OpenAI-compatible endpoint. README states no cloud service and no paid tier. · Free / OSS
- GitHub stars
- 468
- Last release
- 2026-06-12
- Last commit
- 2026-06-12
- First catalogued
- 2026-06-28
Strengths
- Mem0 API-compatible (`add()`/`search()` same shapes) so existing Mem0 code and framework adapters keep working
- Per-character isolated memory profiles for role-play / companion / multi-persona agents, plus LLM semantic-clustering dedup
- Multimodal video memory (frames→captions→vector DB) with ReAct video QA; fully local option via Qwen + FAISS, MCP server included
Watch out
- README reports self-reported benchmark numbers (86.33% on the ZH-4O Chinese role-play set, '19% higher than Mem0') — record via harvest-benchmarks with selfReported:true, not here
- Smaller community (~470 stars) and young (first release 2025-12); the video pipeline needs configured VLM + embedding services to generate artifacts
- Built on / forked from Mem0's OSS client, so some defaults inherit mem0ai provider behavior
Best for
- Teams wanting a local, Mem0-compatible memory layer with strong per-character isolation and optional video memory
Benchmark results
No sourced results yet.
Sources
- TeleMem README (vendor)
- GitHub API repo metadata (stars, Apache-2.0 license, v1.7.1 release) (third-party)
- TeleMem: Building Long-Term and Multimodal Memory for Agentic AI (arXiv) (paper)
Last verified 2026-06-28 · updated by discover-frameworks