Skip to content
ALM Agents
Developer Preview — known limitations are marked in red and will be fixed for production.

Project memory

Agents don’t remember between tasks on their own. Project memory changes that, under your control: when you send work back with a rule worth keeping, the agent proposes a short lesson, you approve it, and from then on it is part of the agents’ instructions.

How a memory is born

Section titled “How a memory is born”
flowchart LR
  F["<b>Send back</b><br/><small>with Teach ticked</small>"]:::you --> P["<b>Proposal</b><br/><small>written by the agent</small>"]:::run
  P --> R{"<b>You review</b><br/><small>edit text · pick agents</small>"}:::you
  R -- approve --> A["<b>Active</b><br/><small>in the agents' instructions</small>"]:::ok
  R -- discard --> X["<b>Retired</b>"]:::mute
  A -- retire --> X
  X -- reinstate --> A
  A --> U["<b>Run record</b><br/><small>lists the memories used</small>"]:::pol
  classDef pol fill:#dcecec,stroke:#0d6e6e,stroke-width:1.2px,color:#12302f
  classDef run fill:#ece5fb,stroke:#7c3aed,stroke-width:1.2px,color:#2b1d4f
  classDef ext fill:#e3ebfb,stroke:#2563eb,stroke-width:1.2px,color:#16264a
  classDef you fill:#fdf0d8,stroke:#b7791f,stroke-width:1.2px,color:#4a3210
  classDef ok fill:#e3f2e7,stroke:#1f7a3d,stroke-width:1.2px,color:#143d22
  classDef bad fill:#fbe4e2,stroke:#b3261e,stroke-width:1.2px,color:#4d1612
  classDef mute fill:#f1f4f6,stroke:#8a96a3,stroke-width:1.2px,color:#2a3440

Review an agent’s proposed memory: edit, choose agents, approve

  1. Send back with Teach. In the decision box, type your feedback and tick 🧠 Teach Atlas from this.

    Summaries for me should end with the requirement id in square brackets, so I can paste them into the review minutes.

  2. The agent fixes the work — and proposes a memory. It turns your specific feedback into a general rule, and decides whether it is about how it works or a convention for the whole team:

    End every requirement summary with the requirement id in square brackets, e.g. [PU-2081].

  3. You review it. 🧠 Review memory ▸ appears in Needs you. Edit the text until it says exactly what you mean, tick which agents get it (or whole team), and Approve — or Discard.
  4. It takes effect. From the next run on, the chosen agents find the lesson in their instructions under “Lessons from your supervisor”.

Who gets a memory

Section titled “Who gets a memory”

A memory’s For agent field lists the agents that get it. “Whole team” means the agents hired at that moment — agents hired later don’t get it automatically. That is deliberate: a new agent starts clean. Add it to a memory if it should.

Memories belong to the project where they were learned. A lesson about your project’s conventions never leaks into another project, even if the same agent works there. Lessons that hold everywhere belong in a skill.

The memory work item

Section titled “The memory work item”

A memory work item with its memory graph

Each memory is a work item of type Agent Memory (aiMemory) in your project:

Field
Status Proposed → Active (approved) → Retired
For agent The agents whose instructions include it
Author The agent that proposed it
Description The lesson
learned from Link to the task where you gave the feedback
Comments “✅ Memory approved for … by …” — the decision on record

Its Memory graph panel shows where it sits: the feedback it was learned from, the agents that have it, and how many runs it shaped.

Manage memories

Section titled “Manage memories”
  • An agent’s ⚙ → Memories shows how many memories are in its instructions, with Show in Polarion ↗ to list them.
  • To stop a memory, open it and use the workflow action Retire. Reinstate brings it back.
  • Many memories make instructions long; the menu suggests tidying up when an agent has more than 15.