Toolkits and memory mode

  • Desktop app
  • All plans
View as Markdown

What it does

A toolkit is a named, reusable bundle of skills, saved connections (SSH servers, databases, dev commands), MCP servers and environment variables, kept in your account library. You attach it to an agent, a single backlog ticket, a team step, a project or a folder. It is attached by reference: editing the toolkit changes every agent that uses it at its next launch, nothing has to be rebuilt.

The memory mode is a separate setting on an agent and on a ticket: Off, Read only, or Read and write. Persona, toolkits and memory are three independent controls: an agent can keep its persona and tools without touching the project memory, and a ticket can get a research toolkit for one run without changing its agent.

Where to find it

  • The library: the "Toolkits" tab of the sidebar on the projects home, next to Projects and Agents (list, + to create, click a toolkit to edit it). The briefcase button next to Agents in the title bar opens that tab (left pill: Agents and Toolkits; right pill: Home and Open space). The "Manage" link of any "Toolkits" picker opens the same library.
  • Agent: Edit agent > Advanced options > Toolkits, and Project memory just below. Same two fields in Add agent > Advanced.
  • Ticket: the ticket window, below the skills ("Toolkits" and "Project memory"). Not shown for a ticket assigned to a team.
  • Team step: the step's agent configuration (Advanced > Toolkits).
  • Project: Project settings > Capabilities > Project toolkits.
  • Folder: Folder settings > Knowledge > Folder toolkits.
  • Phone: Edit agent and New agent sheets, and the ticket detail, with the same fields; "Manage toolkits" opens the library, which works even when the desktop is closed.

How to use it

  1. Open a Toolkits picker and click Manage, then New toolkit.
  2. Give it a name, then pick its skills and connections, paste its MCP servers (same JSON as Settings > MCP servers) and its environment variables (KEY=value lines). Use {{secret:NAME}} for a token.
  3. Save, then attach it where you need it with "Add toolkit".

A ticket's toolkits are added to its agent's own. When the ticket's skills mode is Replace, the run gets only the ticket's toolkits and skills: the agent's, the project's and the folder's are dropped.

Settings

  • toolkitIds (agent, Edit agent > Advanced options > Toolkits): toolkits of this agent. Default none.
  • memoryMode (agent, Edit agent > Advanced options > Project memory): off, read or write. Default unset = inherit the project memory switch and the AgentsRoom tool settings.
  • toolkitIds / memoryMode on a backlog ticket (ticket window): for that run only. Unset memory mode = the agent's own.
  • toolkitIds on a team step (step configuration): the step's toolkits. A team step has no memory mode: use Read only on the step instead.
  • toolkitIds (project, Project settings > Capabilities > Project toolkits): every agent of the project.
  • toolkitIds (folder, Folder settings > Knowledge > Folder toolkits): every agent of every project in the folder; a sub-folder adds to its parents' toolkits.

Agent tools (MCP)

  • toolkits_list: the account's toolkits (ids, names, what they hold, never a secret value).
  • toolkits_get: one toolkit in full.
  • toolkits_save: create or update a toolkit (only on the user's request).
  • toolkits_delete: delete a toolkit (explicit instruction only).
  • agents_save, backlog_create / backlog_update, teams_save and settings_set (project, folder) accept toolkitIds; agents_save and the backlog tools accept memoryMode.

Providers

All providers. Skills reach every CLI as prompt text, and as native skills on Claude Code and Antigravity. MCP servers are written in each CLI's own format at launch. Read only memory hides memory_save and memory_delete and tells the agent the memory is read only; Off hides the memory tools and the memory pointer.

Mobile

Present: pick and create toolkits, set the memory mode on an agent and on a ticket. The phone writes the library directly to the account, so it works with the desktop closed; the desktop applies everything at the next launch.

Limits

  • Changes apply at the agent's next launch, not to a running session.
  • A project skill or a project connection inside a toolkit only applies in its own project.
  • A toolkit belongs to one account: a team shared with teammates keeps the ids, but a teammate's launch only resolves their own toolkits.
  • 20 toolkits per level, 200 toolkits per account.

Common questions

  • Can I reuse the same skills and connections on several agents without rebuilding them? Yes, put them in a toolkit and attach it to each agent.
  • Can I give a ticket an MCP server or a connection without changing its agent? Yes, attach a toolkit to the ticket.
  • Can I stop one agent from writing to the project memory but keep its persona? Yes, set its Project memory to Read only (or Off). The Minimal system prompt is not needed for that.
  • If I edit a toolkit, do agents update? Yes, at their next launch.

Related