🚀 MemGPT Q2 2024 Developer Roadmap

[cpacker]

Q2 2024 Roadmap

🚀 Link to GitHub project board tracking all of the roadmap items
👋 Looking for smaller things to work on? Check the community contributions or bug tracker project boards
✍️ Leave comments or message us on Discord to suggest changes to the roadmap

---

More MemGPT LLM backends / API supported [early April]

• Groq
  • feat: added groq support via local option w/ auth #1203
  • TODO via beta function calling API
    • feat: groq support via official tool-calling API #1257
• Claude
  • feat: Anthropic Claude API support #1239
• Cohere
  • feat: add Cohere API support (Command-R+) #1246
• Gemini
  • feat: add Google AI Gemini Pro support #1209
• Together
• Mistral
• Add unit tests for all the officially supported API providers (with reference configs contained in main) to avoid regression
  • OpenAI, Anthropic, Groq, Together, Mistral, Gemini
• Consolidate inference backend options into a single config style:
  • i.e.: openai-chat-completions (= OpenAI, Azure, APIs with function calling), openai-completions (= vLLM, lmstudio, ollama), ...

MemGPT developer examples

• Native multi-agent interaction (via multiple MemGPT agents running on a MemGPT server process)
  • Example "agent orchestrator" meta-agent where one agent controls send_message calls to other agents
    • in this example, all the agents share a groupchat state similar to AutoGen
  • Example free-chat where each agent can freely broadcast send_message calls to other agents
    • need to handle "race conditions" where agent is busy (the calling agent should receive an informative message reply)
    • in this example, the only shared state between agents is via communication on send_message calls
• Discord / Slack chatbot examples (connecting send_message to external APIs)
  • Example Discord + Slack + Twilio send_message tool + message listen hook (needs a dedicated section in dev portal)
• GitHub / Discord support chatbot examples (e.g. Dosubot)
  • Example read_issue, list_issues + comment-posted-to-API-call hooks

MemGPT server

• One-click deployment container [early April]
• Instructions / tutorial for deploying MemGPT server to Azure/GCP/AWS/... [mid April]

MemGPT API

• Token streaming support [mid-April]
• Production-ready stable API [mid-April]

OpenAI Assistants API

• Continued improvements to OpenAPI Assistants API support: OpenAI Assistant API Compatibility Tracker  #892

MemGPT Client / SDK

• Javascript/Typescript client

Developer portal / Chat UI

• Alpha release [early April]
• Addition of missing features for beta release in MemGPT v0.4 [early April]
  • Preset creation / editing
  • Custom function / tool creation (via the UI)
  • Message (user+system+assistant) editing + rerunning / regenerating messages
  • Custom prompt formatting
• Cron job scheduling inside of dev portal
  • Make it easy to schedule automated jobs that hit the MemGPT server
  • Cron-style custom functions: MemGPT can schedule one off or recurring messages to itself, ie ‘Scheduled Inner Monologue’

Hosted service

• Hosted service [end of April]
  • Release hosted MemGPT server so that developers can directly interact with the API
• Allows developers to use the MemGPT API without requiring any self-hosting (just API keys)
  • Release hosted chat UI app (with guest + login modes) to allow easy use / experimentation with MemGPT via chat UI only
  • Accounts are shared with the hosted API server (allows interacting with the same agents via hosted API + hosted chat UI)

⚡ Streaming (token-level) support

• Add streaming support for CLI interface with OpenAI-compatible endpoints
  • feat: add streaming support for OpenAI-compatible endpoints #1262
• Allow streaming back of POST requests (to MemGPT API / server)
  • feat: add token streaming to the MemGPT API #1280
  • In MemGPT function calling setup, this likely means:
    • Stream back inner thoughts first
    • Then stream back function call
    • Then attempt to parse function call (validate if final full streamed response was OK)

---

Miscellaneous features (Q2+)

👥 Split thread agent

• Support alternate “split-thread” MemGPT agent architecture
• SplitThreadAgent-v0.1 runs two “prompt” / “context” threads
  • DialogueThread that generates conversations and calls utility functions (e.g. run_google_search(...) or call_smart_home(...))
  • MemoryThread that is a passive reader of the ongoing conversation, and is responsible for memory edits, insertion, and search
    • core_memory_replace , core_memory_append
    • archival_memory_search, archival_memory_insert
    • conversation_search, conversation_search_date
    • Question: should these be usable by the DialogueThread too?

🦙 Specialized MemGPT models

• Release (on HuggingFace) and serve (via the free endpoint) models that have been fine-tuned specifically for MemGPT
  • “MemGPT-LM-8x7B-v0.1” (e.g. Mixtral 8x7B fine-tuned on MemGPT data w/ DPO)
  • Goal is to bridge the gap between open models and GPT-4 for MemGPT performance

👁️ Multi-modal support

• Start with gpt-4-vision support first to work out the necessary refactors required
  • Will require modifications to the current data_types stored in the database
• Work backwards to LLaVA

👾 Make MemGPT a better coding assistant

• Coding requires some coding-specific optimizations
  • Better support for generating coding blocks with parsing errors
  • Add specific grammars / model wrappers for coding
• Add support for block-level code execution
  • CodeInterpreter style

📄 Make MemGPT a better document research assistant

• Add more complex out-of-the-box archival_memory_search replacements
  • e.g. using LlamaIndex RAG pipelines

🔧 Better default functions

• E.g. better out-of-the-box internet search

⏱️ Asynchronous tool use support

• Support non-blocking tool use (Feature Request: sending message to agent without user role #1062)
  • E.g. image generation that takes ~10s+ should not block the main conversation thread
  • Implement by returning the "TBD" tool response immediately, then inject full response later

🧠 Better memory systems

• core_memory_v2, archival_memory_v2, etc.
  • e.g. core_memory_v2
    • add more structure (insertions are key, value style only)
  • e.g. archival_memory_v2
    • add more metadata tagging at insertion time
      • type: [memory, feeling, observation, reflection, …]
    • add an asynchronous “memory consolidation” loop
      • every N minutes (or once per day), a task runner starts that tries to consolidate all the archival memories
    • add more structure in the storage
      • not just a vector database
      • knowledge graph?
      • hierarchical storage?

[atljoseph]

Hi, just wanted to say this is an awesome project, and that the ability to have a fully featured ui to get started is really important to me. Nice to see it on the roadmap. That along with non-trivial examples makes it an easy choice to go with memGPT.

[cpacker]

#1533