Letta
Choose a persistent-agent workspace by agent count, memory size and model usage, with personal and developer plans kept distinct.
- Starting price
- Personal Pro USD 20/month
- Free access
- Three stateful agents on Free
- Billing unit
- Personal plan plus usage/model bills
- Platform
- Web app and CLI; separate API plans
- Verified
- 27 Oct 2026
Overview
What you provide
- Configure an agent identity, memory and permitted tools, then provide tasks through the supported app or CLI.
What you get
- The agent returns task responses while retaining state and memory according to its configuration and plan limits.
Choosing this tool
Letta is relevant when an assistant needs an identity and memory across sessions instead of a disposable chat. The personal plan packages that workflow, while developer API automation uses a different usage bill. Compare LangGraph when engineers want to own orchestration and persistence code; Agno offers another code-first agent framework. Those libraries have a different operational burden from a hosted personal workspace.
A decision worked through
Pro costs USD 240 for twelve monthly subscriptions, before paid Letta Auto overage or external model plans. Moving from three Free agents to twenty Pro agents increases the count ceiling by seventeen, but does not create seventeen times more included inference.
Letta alternatives for persistent agents with managed memory
Compare Letta with LangGraph library, Agno framework by the actual task, billing unit and operating responsibility.
| Product | Price / unit | Free allowance | Input → output | Key limit |
|---|---|---|---|---|
| LettaRead profileOfficial source ↗ | Personal Pro USD 20/month | Three agents; external models extra | Configured memory supports ongoing tasks. | Developer API billing is separate. |
| LangGraph libraryOfficial source ↗ | MIT library licence: USD 0 | Library; no hosted quota included | Orchestration code runs stateful agents. | Engineers provide persistence and hosting. |
| Agno frameworkOfficial source ↗ | Apache-2.0 framework licence: USD 0 | Framework; model bills separate | Python code configures agents and teams. | Deployment and runtime operation are yours. |
Setup & workflow
- Create one agent with a narrow purpose and minimal tool access.
- Inspect what the agent retains after a small task and correct unsuitable memory.
- Monitor usage and storage before enabling longer or parallel sessions.
Requirements & installation
- Decide whether this is personal interactive use or a developer API service, and set the model spending boundary.
Limits & review
- The pricing page does not quantify every Letta Auto token allowance. Heavy tool use and generation can consume quota quickly; memory capacity is not a substitute for inference credit. Do not assume the personal Pro plan includes developer API workloads.
Your part
- Review stored memory and tool permissions, and approve consequential actions before allowing autonomous execution.
Sources & verification5
This profile is based on official sources, not a hands-on product test.
Vendor descriptions and demos document advertised features. Editorial guidance is based on these sources.
- Official Letta pricingletta.com
- Official Letta productletta.com
- Letta official referenceletta.com
- LangGraph library official referencegithub.com
- Agno framework official referencegithub.com
