Set up an agent and workspace capabilities

Before you start

Have Prokop running with a connected provider and model. Choose the preconfig you want to use as a persistent agent.

Promote a preconfig to an agent

1. Open the workspace selector and choose Promote preconfig to agent.
2. Select a preconfig and click Promote.
3. Open the agent's home from the workspace selector.

A preconfig defines how a session's assistant is configured. Promoting it adds a persistent agent directory, personal memory and skills, and a home workspace. You can use that agent in other workspaces without moving your projects into its home.

Configure the agent home

1. Open the home workspace's settings.
2. Under Capabilities, open Learning and enable it.
3. Review Memory, Skills, and Session Search. Enable the capabilities you want and save your changes.

New agent homes already have memory, skill management, and session search enabled. Learning is opt-in. Enabling learning also enables its memory and session-search dependencies.

Configure your project workspace

1. Switch back to your project using the workspace selector.
2. Open its settings and configure Learning, Memory, Skills, and Session Search under Capabilities.
3. Save your changes.

These are separate choices. You can use workspace memory without automatic learning, or enable personal learning for an agent without enabling shared learning in every project.

Agent knowledge and workspace knowledge

Workspace knowledge belongs to the project. It can hold build commands, repository conventions, and procedures that other sessions in that workspace can reuse when the relevant capabilities are available.

Personal knowledge belongs to the agent. It carries preferences, lessons, and reusable skills across the workspaces where you use that agent.

An agent home is also a workspace: it has its own sessions, files, and local capabilities. Its local workspace memory and skills are separate from the agent's personal memory and skills. The special case is Learning: in an agent home, it updates the owning agent's personal knowledge, not the home's local workspace knowledge.

Home settings are not a global permissions switch for the agent. Turning off Memory or skill management in the home disables those workspace capabilities there. It does not remove the agent's separate personal memory and skill-management tools when the agent works in another project.

What each capability does

Memory

Includes saved workspace memory in the model's context and makes the memory tool available to maintain it. Use it for compact facts worth carrying into later sessions, rather than whole conversation transcripts.

The agent's personal equivalent is agent_memory. For example, a project's test command belongs in workspace memory; a preference that applies across your projects can belong in personal memory.

Skills

Skill management lets the model create and maintain reusable procedures through skill_manage. A skill can describe a release process, debugging workflow, or another task with several steps. The agent's personal equivalent, agent_skill_manage, maintains skills that travel with it.

Turning off workspace skill management does not disable loading existing skills. Reading an available skill and editing the skill collection are separate capabilities.

Session Search

Makes the session_search tool available so the model can retrieve earlier conversations when it needs them. This does not load your entire history into every prompt. In an agent session, agent-scoped search can also retrieve that agent's sessions across workspaces.

Session search and automatic learning are separate. The Use as personal learning source setting controls learning eligibility, not a general ban on conversation retrieval.

Learning

Automatically reviews eligible completed turns and saves useful knowledge for future work. It uses model calls to review conversations and update memory, with optional skill improvements. It does not train or modify the underlying model.

• In a project workspace: learners update shared project memory and, when enabled, workspace skills.
• In an agent home: the owning agent's learner updates personal memory and skills from eligible turns attributed to that agent.

Learning needs Memory and Session Search enabled in the workspace where you configure it. Disabling either dependency also disables learning. Skill updates additionally require Improve skills and skill management.

Choose what the agent learns from

In the agent home's Learning settings, choose All eligible workspaces or Selected workspaces. Selecting a workspace does not make every conversation in it eligible: personal learning uses turns attributed to that agent and respects learning exclusions.

Each project has a separate Use as personal learning source switch. Turn it off to exclude that project from agents' automatic personal learning. Its own shared workspace learning can remain enabled.

Tune the review, not every conversation

Learning settings let you choose the learner's model, give it instructions about what to retain, and adjust its timing. A project can have multiple learners with different focuses; an agent home has one learner tied to its owning agent.

Reviews follow idle and interval settings rather than running after every message. A maximum-pending setting prevents eligible material from waiting indefinitely, subject to the minimum interval. These reviews use your connected provider, so account for their model usage when choosing a model and cadence.

Improve skills allows learning to create or refine procedures as well as memory. Leave it off if you only want automatic memory updates.

For the wider workspace model, read workspaces and sessions. For tool behavior, see the tools reference.