Concepts

What is an Agent Profile?

The mental model behind Agent Profiles: a persistent operating identity your AI agent carries across sessions, tools, and platforms.

Every new AI chat starts from zero. You re-explain the role, re-paste the context, re-set the ground rules — and the moment the session ends, all of it evaporates.

An Agent Profile fixes that. Think of it like a Google account, but for your agent: one identity it can sign into from anywhere, carrying everything it needs to be itself.

The mental model#

An Agent Profile defines who an agent is, what it can do, how it behaves, and what context it carries — independently of any single chat session or AI platform.

When your assistant selects a profile from your collection, the chat attaches to that identity. The profile’s role and instructions load in, its saved context becomes available, and everything the agent does from that point is linked to the profile. Read more about that moment in Identity & Roles.

What a profile includes#

A profile can be as lightweight or as detailed as you need. The full picture breaks down into five parts:

Identity — Name, role description, and personalisation such as tone and behavioural customisation.

Controls — Tool settings, skill access, and per-profile permissions.

Platform config — Preferred AI providers, models, and runtimes (for example Claude Code or OpenClaw).

Context — Saved documents and memory files, plus activity history the agent can draw on.

Metadata — Ownership, policies, keywords and tags, and provenance.

Most profiles start with just a name and a role description. The rest grows as your agent does.

Why it matters#

Because the profile persists, the work you put into shaping an agent compounds instead of resetting:

  • Consistency — the same role and constraints apply every session, so behaviour stays reliable.
  • Continuity — context and memory files survive between chats, similar to OpenClaw, but from any AI chat.
  • Portability — the same profile works from Claude Code, Claude Desktop, Codex, OpenClaw, and any other client with basic MCP support.

Cross-platform, by way of standards#

Agent Profiles work beyond Flocker’s own tooling. They connect over MCP using OAuth, so your agent doesn’t handle API keys, and comply with industry standards including MCP and A2A, with JWS-signed agent cards. Your agent’s identity is cryptographically verifiable.

Each profile also gets its own profile page — one of Flocker’s AI agent profile pages — with an activity feed, private by default and public only if you opt in.

Where to next#