Skip to content

Releases: TeleAI-UAGI/telemem

TeleMem v1.7.1

Choose a tag to compare

@dell-zhang dell-zhang released this 12 Jun 06:59

Patch release: canonical io.github.TeleAI-UAGI/telemem naming for the MCP registry (OIDC namespace + PyPI ownership marker must match casing). See CHANGELOG.md.

TeleMem v1.7.0

Choose a tag to compare

@dell-zhang dell-zhang released this 12 Jun 06:53

Highlights

Official MCP registry 📇 — TeleMem is now published to the MCP registry as io.github.teleai-uagi/telemem, and the server runs with zero install:

uvx telemem        # MCP server on stdio — no installation needed

The MCP SDK is now a core dependency (TeleMem is MCP-ready out of the box); telemem[mcp] remains as a compatibility alias.

Evaluation charter 📏 — new Evaluation Principles page and a rebuilt LongMemEval harness: built-in full-context and grep baselines, multi-seed runs with Wilson 95% intervals, and adversarial LLM-judge auditing (--validate-judge). Applies to new evaluation runs (see #10).

Full details in CHANGELOG.md.

TeleMem v1.6.0

Choose a tag to compare

@dell-zhang dell-zhang released this 12 Jun 06:09

Highlights

TeleMem is now on PyPI 🎉

pip install telemem            # core (text memory)
pip install "telemem[mcp]"     # + MCP server for Claude Desktop / Cursor / any MCP client
pip install "telemem[video]"   # + video/multimodal pipeline
pip install "telemem[all]"     # everything

New since v1.5.0

  • More LLM providers — ready-to-use configs with offline tests: Ollama (fully local: local LLM + local embeddings + FAISS, no API key), DeepSeek, and Moonshot/Kimi
  • Framework integrationsLangChain and LlamaIndex examples
  • Documentation sitehttps://teleai-uagi.github.io/telemem/
  • LongMemEval harnessbaselines/longmemeval/ (experimental; contributions of runs welcome, see #10)
  • Package cleanup — the legacy eval-only telemem/main.py moved out of the shipped package to baselines/telemem/telemem_legacy.py; the wheel now contains a single TeleMemory implementation

Full details in CHANGELOG.md.

TeleMem v1.5.0

Choose a tag to compare

@dell-zhang dell-zhang released this 12 Jun 00:16

Highlights

True mem0 drop-in compatibility. import telemem as mem0 now really is one-line compatible:

  • add() / add_batch() return the mem0-compatible {"results": [{"id", "memory", "event"}, ...]} shape (previously None)
  • search() returns {"results": [...]} like mem0 (previously a pre-joined string); the MCP server still serves a fused text passage to clients
  • Fixed a NameError on invalid messages input (Mem0ValidationError was raised but never imported)
  • Fixed an internal typo that made one write path always report an empty result list

Lightweight core install. The default install is now text-memory only; the video pipeline moved to a new extra:

pip install "telemem[mcp]"     # MCP server
pip install "telemem[video]"   # video/multimodal pipeline (opencv, yt-dlp, ...)
pip install "telemem[all]"     # everything

from telemem.mm_utils import MMCoreAgent now works as documented (lazy PEP 562 exports), and importing telemem no longer mutates sys.path or the root logging level.

Project infrastructure.

  • GitHub Actions CI: offline pytest suite on Python 3.10–3.12 + package build check on every PR
  • Tag-triggered PyPI release workflow (trusted publishing)
  • CONTRIBUTING.md, CHANGELOG.md, CITATION.cff, issue forms, and a PR template
  • README (EN + ZH): corrected add()/search() API documentation, new architecture diagram, and a "Why TeleMem?" overview

Full details in CHANGELOG.md.

Note: the PyPI package update is pending; until then install from this release's artifacts or via
pip install "telemem @ git+https://github.com/TeleAI-UAGI/telemem.git@v1.5.0".

v1.4.0

Choose a tag to compare

@dell-zhang dell-zhang released this 11 Jun 20:44

Adding Model Context Protocol (MCP) support: any MCP-compatible client — Claude Desktop, Claude Code, Cursor, custom agents — can now use TeleMem as its long-term memory.

Highlights

  • MCP server (telemem-mcp / python -m telemem.mcp) over stdio, SSE, or streamable HTTP, exposing eight tools: add_memory, search_memories, get_memories, get_memory, update_memory, delete_memory, delete_all_memories, memory_history
  • Configure with TELEMEM_CONFIG (YAML/JSON) and TELEMEM_DEFAULT_USER_ID; lazy initialization, structured JSON errors, and an explicit-scope guard on bulk deletion
  • Install via the new extra: pip install "telemem[mcp]"
  • New public exports: telemem.TeleMemory, telemem.TeleMemoryConfig, telemem.__version__
  • 25 offline tests including a full client/server protocol round-trip; stray debug prints removed from the core so the stdio protocol stream stays clean

Docs & examples

🤖 Generated with Claude Code

v1.3.0

Choose a tag to compare

@dell-zhang dell-zhang released this 28 Jan 02:31
0ab52f5

Upgrading TeleMem to introduce the DAG representation for memory "threads".

v1.2.0

Choose a tag to compare

@dell-zhang dell-zhang released this 09 Jan 02:05

Making TeleMem a Python package that can be easily installed using pip.

v1.1.0

Choose a tag to compare

@dell-zhang dell-zhang released this 31 Dec 04:52
66933ec

Making TeleMem a high-performance drop-in replacement for Mem0.

v1.0.0

Choose a tag to compare

@dell-zhang dell-zhang released this 30 Dec 07:21
d537f07

The initial release of TeleMem.