Why it matters: MongoDB's document model is a natural fit for session transcript storage — each session is an arbitrarily nested JSON structure (tool calls, message turns, metadata) that maps directly to a BSON document without schema migration or an ORM layer. This positions MongoDB as the recommended production backend for Agent SDK deployments, alongside existing S3, Redis, and Postgres reference adapters.

Status in Aha: Defining Requirements (this is new work — not yet built). Needs Engineering and other cross-functional work (docs, LRA, etc.) before it can ship.

Related links:

• Aha feature: AI-82
• LRA ticket: WRITING-39486
• Reference: MongoDB SessionStore reference adapter PR (#1014, open) confirms buildability
• Reference: S3/Redis/Postgres reference adapters (merged, PR #842) establish the competitive baseline to match
• Reference: PyMongo driver docs — recommended async-capable driver (pymongo >= 4.9 with AsyncMongoClient)

Scope:

• Implement a MongoDB-backed Session/SessionStore adapter for the OpenAI Agents SDK, in both Python and JavaScript
• Align with the SDK's existing Session protocol so it's a drop-in alternative to S3/Redis/Postgres adapters
• Get the adapter merged as a first-party reference example in the SDK's examples directory (matching S3/Redis/Postgres precedent)
• Coordinate with docs, DevRel, and LRA process ahead of launch

Acceptance Criteria:

• [ ] MongoDB SessionStore adapter implemented and passing SDK's Session protocol conformance (Python)
• [ ] MongoDB SessionStore adapter implemented and passing SDK's Session protocol conformance (JavaScript)
• [ ] Adapter supports core session operations: create/read/update/query session transcripts (tool calls, message turns, metadata)
• [ ] Adapter contributed upstream and merged into the OpenAI Agents SDK examples directory
• [ ] Documentation/tutorial published covering setup and usage with MongoDB Atlas
• [ ] LRA (WRITING-39486) completed and signed off
• [ ] Feature ready to ship ahead of the 09/24/2026 release target