ADoor vs Letta
Letta is a platform for stateful agents that keep and edit their own memory. ADoor, your data account, keeps the records those agents share with people and other tools, with access per agent, list, field and entry.
Where ADoor stands out
Shared beyond one agent
A Letta agent keeps its own memory. ADoor lists are shared by people, apps and assistants like ChatGPT.
Access per agent
In ADoor each agent is capped at the person who connected it, and can be narrowed to one list or field.
Approvals
In ADoor a new agent or device waits for your approval.
The products
Letta
Letta runs stateful agents with persistent memory, formerly MemGPT.
- Assistants
- A hosted MCP server at api.letta.com/mcp with a Letta API key
- Plans
- Free plans are limited to 3 stateful agents. Pro is $20 a month for up to 20.
ADoor
ADoor is your data account. You keep lists and tables in a workspace you own, personal or team: a reading list, an incident log, a list of customers. Describe what you want to keep track of and it's set up for you, with real Postgres underneath. Every person, app, assistant, device and job gets its own access.
- Assistants
- ChatGPT, Claude, Grok and Cursor connect through MCP with a sign-in.
- Apps and scripts
- An HTTP API with keys, running the same actions assistants use
- Access
- Per space, list or field, or just the entries where a field equals a value you choose. Reading, adding, editing and deleting entries, and changing a list's fields, are separate permissions. An assistant never gets more than the person who connected it.
- Approvals
- Agents ask for more access and you approve or deny. A new device signs in only after you approve it and choose its access.
- CI jobs
- GitHub Actions signs in with short-lived access, 15 minutes by default and up to 30 days, with no stored secret.
- Revoking
- Takes effect on the next request. What was already saved stays.
- History
- A history of changes, with who made each one and the before and after
- Underneath
- Standard Postgres
- Plans
- Free for 3 tables and 100 rows. Pro is $10 a month per workspace, with 1 GB of storage.
Pricing
| Topic | ADoor | Letta |
|---|---|---|
| Free plan | 3 tables and 100 rows, with a few MB of storage | Free, up to 3 stateful agents |
| Paid plans | Pro: $10 a month per workspace, 1 GB of storage, no per-GB charges | Pro $20 a month, up to 20 stateful agents |
Latency
| Topic | ADoor | Letta |
|---|---|---|
| Published figure | Not published | Not published |
ADoor doesn't publish a latency figure, and Letta doesn't publish one we could find. Speed depends on your network and on what you ask for.
Connectors and API
| Topic | ADoor | Letta |
|---|---|---|
| MCP server | Hosted, with OAuth sign-in. ChatGPT, Claude, Grok and Cursor. | Hosted at api.letta.com/mcp |
| HTTP API | HTTP API with keys: read, add, edit, delete, and change lists and fields | Letta API |
| SQL access | Standard Postgres SQL through MCP and the API, within your access | No |
| Access per agent | Its own access per assistant, app, device and job, down to specific fields, or just the entries that match a value you choose | Not published |
| OAuth or tokens | OAuth for assistants, keys for apps, short-lived sign-in for GitHub Actions | API key |
| SDKs | Coming soon | Not published |
| Webhooks | No | Not published |
| Import and export | Export to CSV or JSON. Import by attaching CSV or JSON in an assistant. | Not published |
ADoor has no sync connectors and no file storage.
Which one should you use?
Use Letta when
you're building agents that learn and keep their own memory over time.
- Stateful agents. Agents keep and edit their memory.
- Open source roots. Letta grew out of MemGPT.
- Hosted MCP. Use Letta from MCP clients.
Use ADoor when
your agents and people need the same records, each with narrow access.
Technical
- A Letta agent writes each lead it qualifies to an add-only ADoor list through the HTTP API, and sales reads it.
- A GitHub Actions job updates a Tools list with 15 minutes of access.
- A Cursor agent reads the same Customers list your Letta agents write to, and asks before it edits anything.
Personal
- A group trip budget: who paid for what and who owes whom. Friends add their own expenses (one grant per person).
- Your car's service history and next due dates, read by ChatGPT when you ask what's coming up.
Team
- Research agents add sources to a list that analysts edit.
- A history of changes shows which agent changed which entry.
- Operations reviews new entries from agents in the history of changes, with the before and after for each edit.
Business
- Each customer's agent sees only that customer's entries.
- A partner's agent reads one list until you revoke it.
- A logistics company's dispatch agent adds delivery exceptions to a list that drivers read on their phones and managers edit.
Questions people ask
Can a Letta agent use ADoor?
Yes, through the HTTP API with a key, within that key's access.
Does ADoor run agents?
No. ADoor keeps the records agents use.
Which assistants work with ADoor?
ChatGPT, Claude, Grok and Cursor connect through MCP with a sign-in. You approve each one and choose what it can reach.
Start
More comparisons
Checked October 20, 2026. Sources: letta.com/pricing, Letta docs, Letta hosted MCP.