Letta memory blocks
Letta documents memory blocks as persistent sections of agent context.
Available now Managed source memoryMCP entry pointInspectable recall evidence
Shipping next Org-shared memory after per-doc ACL
Fig. 2 visible html / no hidden accordion
FIG. 2
Comparison table
The table is the citation surface: concise rows, explicit concessions, and visible cell states. Roadmap-only capabilities render as Blueprint chips instead of available claims.
| Decision point | Answer Engine | Letta |
|---|---|---|
| Primary abstraction What are you buying? | yes A managed memory layer that agents and developers call through MCP. | yes A stateful agent platform centered on memory blocks and agent state. |
| Source inspection Can a reviewer inspect recalled evidence? | yes Source evidence is the lead wedge. | partial Blocks are inspectable, but the model is memory-in-context rather than source-lineage procurement. |
| Multi-source memory How does memory span external sources? | yes Positioned around connected sources, tenant lineage, and MCP workflow context. | partial Strong tools, files, and block model; source breadth depends on the application design. |
| Portability Can memory move between systems? | yes Open MCP surface and source-level evidence reduce lock-in. | yes Letta exposes agent export and import flows. |
| Managed engine Who owns the memory substrate? | yes Answer Engine is the managed memory substrate. | partial Letta can be used through its platform or self-hosted agent stack depending on buyer choice. |
Fig. 3 official sources / current as of 2026.06
FIG. 3
Source notes
Competitor claims link to official public sources where possible. Build-vs-buy figures are Answer Engine planning anchors, not measured product results.
Letta AgentFile
Letta documents import and export of agents through its .af AgentFile format.
Fig. 4 honest concession / no FUD
FIG. 4
Where they win
A credible comparison page states the competitor's best-fit case directly, then narrows the reason to choose Answer Engine.
CONCEDE FIRST Letta wins when you want the agent runtime itself: memory blocks always visible in context, agent state as a first-class object, and exportable agent files.
Fig. 5 differentiators / source evidence
FIG. 5
Why Answer Engine
The differentiators stay tied to inspectability, forgetting, portability, source breadth, and tenant boundaries.
D.01 Available
Source-oriented memory
Answer Engine starts from evidence, source lineage, and tenant boundaries rather than agent state alone.
D.02 Available
Managed substrate
The product is the memory layer that agents call, not the broader runtime where the agent is authored.
D.03 Available
MCP-first integration
Developer adoption starts through the MCP path before the team standardizes agent runtime choices.
Fig. 6 FAQPage schema / crawlable answers
FIG. 6
FAQ
These answers are rendered as plain HTML and mirrored into FAQPage JSON-LD for answer engines.
Is Answer Engine a Letta replacement?
Not exactly. Letta is a stateful agent platform. Answer Engine is a managed memory layer for agents and developers who need source inspection, forgetting, and tenant-isolated recall.
Where does Letta win?
Letta wins when memory blocks, agent state, and the agent runtime are the product surface you want to build around.
Why choose Answer Engine instead of Letta?
Choose Answer Engine when the memory layer should be independent of the agent runtime and should expose source evidence through an MCP-first workflow.
FIG. 7 - DECIDE
Choose the memory layer by the evidence you need to inspect.
Start with Answer Engine when the buying motion depends on inspectable recall, deletion controls, source breadth, and tenant separation.
Use Answer Engine for: managed source memory
Use Letta for: stateful agent runtime
Shared status: org-shared waits for per-doc ACL