TeleMem · AI Nav City

Core facts

Read by a script from the official API, not rewritten by a model.

Repository: https://github.com/TeleAI-UAGI/telemem
License: Apache-2.0
Languages: Python
Maintainer: TeleAI-UAGI
Homepage: https://teleai-uagi.github.io/telemem/
Verified: 2026-11-01

What it is

TeleMem is an agent memory management layer. The project's title is "TeleMem: Building Long-Term and Multimodal Memory for Agentic AI". TeleMem provides a pipeline of video frame extraction, caption generation and vector database construction. The ultimate goal of the TeleMem project is to use an agent's hindsight to improve its foresight.

Main features

The automatic memory extraction feature extracts and structures key facts from dialogues.
The semantic clustering and deduplication feature uses LLMs to semantically merge similar memories.
Character-profiled memory management builds independent memory archives for each character in a dialogue.
The asynchronous writing feature employs a buffer and batch-flush mechanism for persistence.
Semantic retrieval combines FAISS and JSON dual storage.
TeleMem can query video content using natural language, supporting ReAct-style multi-step reasoning.

Getting started

TeleMem is available on PyPI.
To use the repository's local Qwen + FAISS configuration, the configuration file is loaded explicitly.
TeleMem supports MiniMax as an LLM backend via its OpenAI-compatible API.
The add method injects one or more dialogue turns into the memory system.

From the README · shell

pip install telemem            # core (text memory)
pip install "telemem[mcp]"     # + MCP server
pip install "telemem[video]"   # + video/multimodal pipeline
pip install "telemem[all]"     # everything

From the README · shell

uv sync --all-extras   # install TeleMem (editable) + all extras, incl. MCP
uv run python examples/quickstart.py

From the README · shell

export OPENAI_API_KEY="your-openai-api-key"

From the README · python

# python examples/quickstart.py
import telemem as mem0

memory = mem0.Memory()

messages = [
    {"role": "user", "content": "Jordan, did you take the subway to work again today?"},
    {"role": "assistant", "content": "Yes, James. The subway is much faster than driving. I leave at 7 o'clock and it's just not crowded."},
    {"role": "user", "content": "Jordan, I want to try taking the subway too. Can you tell me which station is closest?"},
    {"role": "assistant", "content": "Of course, James. You take Line 2 to Civic Center Station, exit from Exit A, and walk 5 minutes to the company."}
]

memory.add(messages=messages, user_id="Jordan")
results = memory.search("What transportation did Jordan use to go to work today?", user_id="Jordan")
for hit in results["results"]:   # same result shape as mem0
    print(hit["memory"])

From the README · python

from telemem.utils import load_config
import telemem as mem0

config = load_config("config/config.yaml")
memory = mem0.Memory(config=config)

From the README · shell

export MINIMAX_API_KEY="your-minimax-api-key"
export OPENAI_API_KEY="your-openai-api-key"  # still needed for embeddings

Who it is for

TeleMem applies to multi-character virtual agent systems. TeleMem also applies to video content QA and reasoning. MCP-compatible clients such as Claude Desktop, Claude Code, Cursor and custom agents can use TeleMem as long-term memory.

The MCP server exposes eight tools. Destructive bulk deletion on the MCP server always requires an explicit scope. TeleMem itself collects no telemetry.

TeleMem disables anonymized PostHog usage telemetry by default. The search memory function performs semantic vector-based retrieval of relevant memories with context-aware recall. The limit parameter of the search method sets the max number of results, with a default of 100.

The frame extraction stage decodes video to JPEG frames at configured FPS. The vector database construction stage generates embeddings for semantic retrieval. The max_iterations parameter of multimodal search sets the maximum MMCoreAgent reasoning iterations, with a default of 15.

Multimodal search internally uses MMCoreAgent, employing a THINK, ACTION, OBSERVATION loop with three specialized tools. During multimodal memory adding, if the target file for a stage already exists, that stage is automatically skipped to save computational resources. TeleMem automatically creates a structured text memory storage layout organized by session and character.

TeleMem is licensed under the Apache 2.0 License. CI runs the offline test suite on Python 3.10–3.12 for every PR. TeleMem is made by Bloo-Mind AI Ltd and the Ubiquitous AGI team at TeleAI.

Related tools listed on this site.