Letta

  •  Letta Inc., SF, spun out of UC Berkeley BAIR MemGPT paper, $10M seed (Felicis/Founders Fund/YC, Sep 2024), ~23K+ GitHub stars, Apache-2.0. Renamed MemGPT→Letta in 0.5.0 (mid-2024); old memgpt PyPI package unmaintained. No acquisition, no sunset. Category ai-productivity-data, pipeline_stage=none. File under Dev​ (agent-memory/runtime sublane, sibling to LangGraph/LangSmith/Braintrust).Official site: https://www.letta.com​

    Pricing: https://www.letta.com/pricing

    Docs: https://docs.letta.com

    GitHub: https://github.com/letta-ai/letta

    Letta — Memory-Native Stateful Agent Runtime with Tiered Self-Managing Memory (Core / Recall / Archival + Letta Code)

    Snapshot

    Field Value
    Status Active (changelog Aug 2026: 128K ctx, skill learning, Conversations API, Letta Code desktop Apr 2026, Channels May 2026)
    Origin UC Berkeley BAIR MemGPT paper → Letta Inc. SF 2024, Charles Packer/Sarah Wooders/Vivian Myers
    Lane Stateful agent runtime + memory OS (not eval-only, not observability-only, not pure coding agent)
    Core idea Agent manages its own memory tiers via tool calls (OS-style RAM/disk metaphor)
    Center of gravity 2026 Letta Code (memory-first coding agent, #1 model-agnostic OSS on Terminal-Bench 42.5%)
    Category ai-productivity-data / Dev (agent-memory sublane)

    Features (table)

    Layer Capability Notes
    Memory tiers Core​ (in-context blocks, editable by agent), Recall​ (searchable conversation history in DB), Archival​ (vector store, unlimited cold storage) Agent decides read/write/evict via tool calls; survives restart
    Agent loop Tool-calling loop, self-editing memory, context windowing, sleep-time consolidation Not ReAct-only; MemGPT-style virtual context
    Letta Code Memory-first coding agent, desktop app (mac/win/linux Apr 2026), Channels (Slack/Telegram/WhatsApp/Discord/Signal May 2026), Terminal-Bench SOTA Uses Letta Auto (managed model routing)
    Skill learning (Dec 2025) Agents extract reusable skills from trajectories, store in memory, reuse later (+21.1% TB1, +36.8% TB2) Online learning, no fine-tune
    Conversations API Parallel conversations share agent memory; MemFS memory filesystem Git-backed memory (late 2025): version/diff/rollback
    Tools / MCP Python tool bindings, remote MCP server support, scheduled runs, subagents, secrets mgmt Client-side/remote MCP no server-exec charge
    Surfaces REST API (v1 agents/messages/tools/blocks/archives), Python SDK letta-client, TS SDK @letta-ai/letta-client, CLI, ADE (web UI), desktop app Self-host parity with cloud core
    Deploy Self-host (Docker, Postgres/SQLite/Redis/ClickHouse/Turbopuffer), Letta Cloud sandboxes, BYOK all tiers Apache-2.0 core; platform closed cloud
    Enterprise SAML/OIDC SSO, RBAC, increased quotas, private model deploy, dedicated support No native APM/eval (pair with Langfuse/Braintrust)
    No No eval/CI-gate suite (≠ Braintrust/LangSmith), no drag-drop builder, no non-LLM APM Scope = agent runtime+memory

    Pricing (2026-09, USD)

    Tier Platform fee Includes / metering
    OSS Self-host​ $0​ (Apache 2.0) pip install letta[server], BYOK, pay LLM vendor only, no agent cap
    Cloud Free​ $0​ ≤3 managed stateful agents, BYOK, limited Letta Auto, ADE access
    Pro​ $20/mo​ ≤20 stateful agents, Letta Auto weekly+monthly quota, remote sandboxes, pay-as-you-go overage, image-gen credits
    Max Lite​ $100/mo​ ≤50 stateful agents, higher Auto quota
    Max​ $200/mo​ Highest personal quota, early-access features, not for commercial deploy
    API / Developer​ $20/mo base​ Unlimited agents + 0.00015/sec server-side tool exec​ + pass-through LLM tokens; client/remote-MCP tools free
    Teams Pro​ $20/seat/mo​ Shared org agents, access control, Pro features per seat
    Enterprise​ Custom SAML/OIDC, RBAC, private models, volume pricing, SLA

    starting_price=$0; first paid $20/mo (Pro or API base). LLM token cost always billed by your provider (BYOK).

    Reviews (2026)

    Source Signal
    AI Pedia / AI Tools Atlas / AITrendTool 7.5–8/10; “memory-as-infra done right”, agent self-edits core blocks, Letta Code Terminal-Bench SOTA, Apache-2.0 self-host. Criticized: 3-agent free cap tight, memory governance (retention/deletion/sensitive data) is dev responsibility, no built-in evals, Max tier confusing for commercial use
    Community (GitHub 23K★, Reddit/Discord) Loved: crosses model swaps without losing state, MCP + Channels, skill learning. Disliked: learning curve on memory-block design, Postgres ops for self-host, Letta Auto overage surprise, not turnkey for no-code PMs
    Aggregate Best stateful-memory runtime for agents that must remember users/projects across weeks; wrong for one-shot stateless calls or teams wanting LangSmith-style eval gates out of box

    Access / Integration (table)

    Method Supported
    SDK Python letta-client, TypeScript @letta-ai/letta-client
    REST v1 /agents /messages /tools /blocks /archives /runs /mcp etc.
    CLI letta server, letta run, letta create-agent
    Self-host Docker, Postgres/SQLite, Redis, ClickHouse (traces), Turbopuffer (tool search)
    Channels Slack, Telegram, Discord, WhatsApp, Signal (Letta Code)
    MCP Exposes MCP server; consumes remote MCP tools
    Models OpenAI, Anthropic, Mistral, Gemini, DeepSeek, Qwen, Ollama/LM Studio (BYOK)
    No No webhook-only SaaS for non-devs, no REST “score text” endpoint

    Best For / Not For

    Best For Not For
    Agents that must remember users/preferences/project state across sessions Stateless one-shot LLM calls (→ raw SDK)
    Long-running assistants (customer success, tutoring, career coach) Teams wanting built-in evals/CI gates (→ Braintrust/LangSmith)
    Model-agnostic stacks swapping GPT↔Claude↔Qwen without losing state No-code drag-drop builders (→ Lindy/n8n)
    Memory-first coding agents (Letta Code learns repo over time) Simple RAG+vector-store needs (→ Mem0/Zep cheaper)
    Self-host OSS with Postgres-backed agent state LangChain-locked teams wanting zero new runtime (→ LangGraph)

    Competitors (lane map)

    Lane Tool Letta edge / gap
    Agent runtime LangGraph​ LangGraph = graph orchestration + deploy (LangSmith); Letta = memory-OS + self-editing agents, model-agnostic
    Agent framework OpenAI Agents SDK, Microsoft Agent Framework​ Vendor/eco-locked; Letta Apache-2.0 + BYOK + tiered memory
    Memory-layer add-on Mem0, Zep​ Mem0/Zep bolt onto existing agents; Letta is full agent runtime with memory native
    OSS observability Langfuse​ Langfuse traces/evals; Letta runs agents, pairs with Langfuse for obs
    Eval/obs platforms Braintrust, LangSmith​ They observe/eval; Letta executes stateful agents (complementary, not direct)
    Coding agent Cline, Aider, Claude Code​ Those are coding agents; Letta Code is memory-first variant with Letta runtime underneath
    In-list overlap Exa, Cody, JB AI, Continue, Aider, Cline, Mintlify, Braintrust, LangSmith, Q Dev(sunset), Phind(shutdown) Dev lane adjacent (agent-memory vs coding/eval/docs)
  • Tags / Keywordsletta,active-2026-09,formerly-memgpt,stateful-agent-runtime,agent-memory-os,core-recall-archival-memory,self-editing-memory,letta-code,terminal-bench-sota,mcp-server,channels-slack-telegram-whatsapp,skill-learning,apache-2.0-self-host,pro-20-mo,api-20-base-0.10-agent,uc-berkeley-bair,10m-seed-felicis,not-shutdown,ai-productivity-data,pipeline_stage=none,competitors-langgraph-mem0-zep-braintrust-langsmith,dev-agent-memory-sublane

     

Do Not Sell or Share My Personal Information Cookie Settings