openai / openai/openai-agents-python

Feature: memory portability across handoffs — verified memory bundles for agents and sessions

Open
#5,089 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
29.6k
Forks
4.8k
Avg merge
1d 20h
Merged PRs (30d)
123

Description

Maintainer of cognicore-env (open-source, MIT, MCP memory server) — feature proposal with code on offer.

In the Agents SDK, sessions manage conversation history and handoffs pass control — but not knowledge. An agent that learned something verified (a tool's quirks, a customer's constraints, a domain fact) cannot hand that to the next agent except by stuffing it into the prompt, which loses provenance, state, and expiry.

Proposal: memory export/import alongside handoffs. The bundle format we built carries each memory with proof — the command that was run, result, exit code, output hash — plus a chain of custody and trust decay across unverified hops. A memory imported by the receiving agent starts in an observed state and is promoted only after local re-verification. The design includes validity intervals (a fact verified in March is not evidence about April) and owner-per-concept for conflict resolution, from a public multi-team design review.

Concrete case inside this SDK: a triage agent hands off to a specialist agent and passes its verified memory of the issue so far — with the evidence trail — instead of a summarized prompt. The specialist does not re-ask the user what the triage agent already verified.

Our transfer machinery is implemented and tested across four platforms. Happy to contribute an examples/ integration if maintainers see the gap the same way.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reviewing the SDK's session and handoff machinery, then compare its current behavior with the proposed cognicore-env memory bundle design. The payload mentions an examples/ integration as a possible contribution, but maintainers must first define the supported transfer format, verification behavior, and acceptance tests before completion can be measured.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
ai, backend-api-design
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.