Agents
Pasture Protocol is a team of agents: named personas in one local runtime, each with its own identity, selected skills, routing, and context.
The big picture
| Concept | In plain terms |
|---|---|
| Agent | A specialist persona (e.g. main, writer, backend-bot). |
| main | Default coordinator. Always exists; root of the team tree. |
| Skills | Markdown instructions plus JavaScript executors an agent may use, such as browse, memory, github, cron, or agent-send. |
| Identity | Markdown files such as SOUL, WhoAmI, MyHuman, and group context that ground the agent. |
| Agent team | Roster, map, inbox/outbox, live context, missions, tasks, and activity in the dashboard. |
Why multiple agents?
- Separation of concerns: a coding agent with repo access, a writer without sensitive tools.
- Clear personas: different tone, expertise, and memory per role.
- Coordination: main orchestrates specialists via internal messaging.
- Routing: WhatsApp/Telegram groups can be pinned to a specific agent.
What happens on each message
- A message arrives (dashboard, WhatsApp, or Telegram).
- Pasture Protocol picks the routed agent for that chat or group.
- Prompt classifiers choose intent, work mode, skill context, and possible delegation.
- The agent runs a turn: LLM plus optional tool calls through allowed skill executors.
- If needed, it delegates to another agent and synthesizes the reply.
- You see one coherent answer. Delegation stays internal.
Common agent setups
| Setup | Why you would use it |
|---|---|
| main + writer | Main coordinates your request while the writer keeps brand voice, drafts, summaries, and marketing memory separate. |
| main + backend-bot + reviewer | Backend proposes implementation details, reviewer checks risk, and main combines both into one plan or PR response. |
| personal assistant agent | A focused agent can own calendar, Gmail, reminders, and memory without also having code or server tools. |
| group-specific agent | A WhatsApp or Telegram group can talk to a dedicated persona with a limited skill set and group-safe boundaries. |
Agent workspace
Each agent has a dedicated workspace panel in the dashboard with four tabs:
| Tab | What it shows |
|---|---|
| Overview | The agent's name, avatar, role description, enabled skills, and last active time. |
| Context | Current identity files (SOUL, WhoAmI, MyHuman), active mission context, and project notes loaded for this agent. |
| Inbox | Incoming delegation messages this agent has received from other agents via agent-send. |
| Outbox | Delegations this agent has sent to other agents, with their status and any returned results. |
The workspace panel is accessible from the agent roster card. Use it to review what context an agent is working with, trace delegation chains, or confirm that the right identity and mission data is loaded before a session.
Agent avatars
Each agent automatically gets a unique AI-generated portrait. Avatars are generated in the background based on the agent's name and persona description. They appear on roster cards, workspace panels, and in the Reports view. Regeneration happens if the agent configuration changes.