Agent Mode
Create persistent AI teammates with a brief, durable memory, reusable skills, ongoing chats, and access you control.
Agent Mode organizes work around persistent AI teammates. Each agent has a role, its own history and working context, and the ability to improve how it handles repeated work.
The agent is the teammate. Claude Code or Codex can power its chats, but the engine does not define the teammate's role.
What an Agent Contains
| Part | Purpose |
|---|---|
| Name | A recognizable identity in your roster |
| Brief | The job, standards, tone, and boundaries you define |
| Memory | Durable facts and context the agent records for future work |
| Skills | Reusable procedures the agent can follow and improve |
| Places | Folders you explicitly approve for the agent's work |
| Chats | Separate conversation threads with the same teammate |
Create an Agent
- Open Agent from the title-bar mode switcher.
- Select New Agent.
- Give the teammate a specific name.
- Write a brief that defines what it owns and what good work looks like.
- Choose an installed engine. Claude Code and Codex support chat; other detected engines can run as terminal sessions.
- Save, then start the first chat from the agent's page.
BridgeMind One does not bundle either engine. Install the corresponding command-line app first so it can be detected on your Mac.
You can also describe the teammate you need and use Create with AI to draft its name and brief. Review the result before saving it.
Write a Useful Brief
A strong brief answers four questions:
- What do you own? Define the recurring responsibility.
- What context matters? Name the audience, product, or working style.
- What standards apply? State quality, tone, safety, and verification expectations.
- When should you stop? Identify actions that still require builder approval.
Example:
You are our release-notes editor. Turn merged work into concise builder-facing notes. Verify every feature claim against the repository, group changes by outcome, and never publish without approval.
Memory
Memory gives an agent continuity across chats. It is for durable context: preferences, product decisions, recurring constraints, and lessons that should still matter next week.
Use the agent's Memory view to inspect what it knows. Correct stale entries when reality changes. Do not use memory as a substitute for task-specific instructions or as a place to store secrets.
Skills
Skills capture repeatable procedures. An agent can use them to preserve a workflow it has learned instead of solving the same process from scratch each time.
A useful skill should say when it applies, what steps to follow, what to verify, and when to ask for approval. See Skills and Plugins for the full model.
Places and Authority
An agent works from its own home and can be granted additional folders through Places. Grant only the folders the teammate needs.
Plugins, agent messaging, and routine scheduling are separate permissions. Keeping them explicit limits the effect of a mistaken instruction and makes each teammate's authority easier to understand.
Agent Chats
Each chat is a thread with the same agent identity. Start a new chat when the goal changes materially; keep an existing chat when the next request depends on its recent discussion.
When agent messaging is enabled, approved teammates can send each other bounded messages. The exchange appears in both agents' chats and can wake an idle teammate into a fresh chat, subject to the app's safety limits.
Agent Settings
From an agent's Settings view, you can:
- Update its name and brief
- Change the default engine and conversation settings
- Manage approved places
- Choose a memory budget and reflection mode
- Enable or disable messaging and routine permissions
- Delete the agent after reviewing the confirmation
Changing an engine affects future work. Review active chats before switching so you understand which context carries forward.