Verify before you commit
Use this page to sanity-check Letta quickly. We surface the evidence tier, freshness, and failure posture here, then put the official links where you can actually act on them, especially on mobile.
Evidence
Assessed
Docs reviewed · Mar 24, 2026
Freshness
Updated 2026-03-24T23:40:17.91+00:00
Mar 24, 2026
Failures
Clear
No active failures listed
| Dimension | Score | Bar |
|---|---|---|
| Execution Score Measures reliability, idempotency, error ergonomics, latency distribution, and schema stability. | 8.2 | |
| Access Readiness Score Measures how easily an agent can onboard, authenticate, and start using this service autonomously. | 7.8 | |
| Aggregate AN Score Composite score: 70% execution + 30% access readiness. | 8.1 | |
No active failure modes reported.
Published review summaries with trust provenance attached to each card.
Docs-backed Built from public docs and product materials.
Test-backed Backed by guided testing or evaluator-run checks.
Runtime-verified Verified from authenticated runtime evidence.
API key auth: Authorization: Bearer {api_key} header. Keys from Letta Cloud dashboard. HTTPS enforced. No OAuth2. Self-hosted: no API key for local; optional API key for hosted access. Agent state isolated per agent ID.
Letta (formerly MemGPT) addresses the context-window problem through an agent-native memory architecture — agents maintain in-context memory (core memory), recall memory (archival/searchable), and working memory as first-class constructs the agent itself manages. This enables agents to function effectively over very long interactions without context truncation. For agent builders: deploy a Letta agent, send messages, and the agent persists state between calls. Multi-agent support for orchestrated workflows. Confidence is docs-derived.
REST API at letta.ai/v1 (Cloud) or self-hosted. Resources: agents, messages, memory, tools. POST /v1/agents creates a new agent (name, model, persona, human, tools). POST /v1/agents/{id}/messages sends a message to an agent (returns response + tool calls + memory updates). GET /v1/agents/{id}/memory retrieves current agent memory state. POST /v1/agents/{id}/memory/blocks updates a memory block. Python SDK: client.agents.create(), client.agents.messages.create(). WebSocket streaming available.
Agent execution: synchronous response per message turn. Memory updates are atomic per turn. Self-hosted reliability depends on backing database (SQLite default; PostgreSQL for production). Letta Cloud uptime at status.letta.com. Tool-calling failures surface in tool call results within agent response. Long-running tool execution may hit timeouts — configure per-tool timeouts appropriately.
docs.letta.com covers quickstart, agent creation, memory architecture concepts, tool library, multi-agent patterns, and self-hosting guide. Getting started: install letta Python package, run letta quickstart, first agent in under 10 minutes. Active open-source community (GitHub: letta-ai/letta). Letta Cloud for hosted deployments. Research origins from UC Berkeley.
Trust shortcuts
This score is documentation-derived. Treat it as a docs-based evaluation of API design, auth, error handling, and documentation quality.
Read how the score works, how disputes are handled, and how Rhumb scored itself before launch.
Overall tier
8.1 / 10.0
No alternatives captured yet.