Skip to content

Agent Settings

Every agent in Pinchy is configured through a settings page. To open it, navigate to an agent’s chat and click the gear icon in the top bar. Which tabs you see depends on your role and the agent type.

TabWhat it controlsWho can see it
GeneralName, tagline, model, agent typeEveryone who can edit
PersonalityAvatar, personality preset, SOUL.mdEveryone who can edit
InstructionsAGENTS.md operating instructionsEveryone who can edit
PermissionsTool allow-list, directory accessAdmins only, shared agents
Access EnterpriseVisibility mode, group assignmentAdmins only, shared agents
TelegramConnect a Telegram bot to this agentAdmins only

The Permissions and Access tabs exist only for shared agents. Personal agents (the auto-created Smithers each user gets) don’t show them because there’s nothing to share. Admins still see the Telegram tab on Smithers, but the main Smithers bot is managed centrally via Settings → Telegram rather than per-agent.

Agent Settings — General tab

The General tab controls the agent’s identity and the AI model it uses.

FieldDescription
NameThe agent’s display name, shown in the sidebar and chat header. Maximum 50 characters.
TaglineA short description shown below the agent name in the sidebar. Optional.
ModelThe AI model the agent uses, grouped by configured provider (e.g. OpenAI, Anthropic, Ollama).

Below the form fields, the tab shows the agent type — either “Personal agent” or “Shared agent.” Personal agents are private to their owner with isolated memory and conversations. Shared agents are accessible to all team members, and memory from all conversations is shared.

The Danger Zone at the bottom of the General tab lets admins delete shared agents. Personal agents cannot be deleted.

Agent Settings — Personality tab with preset picker

The Personality tab controls how the agent looks and communicates.

Each agent gets a randomly generated avatar. Click Re-roll to generate a new one. The built-in Smithers agent has a fixed avatar that cannot be changed.

Presets define an agent’s tone, communication style, and character. Selecting a preset fills in the SOUL.md file with a matching personality definition. Four presets are available:

PresetSuggested nameStyle
The ButlerSmithersFormal, competent, dry humor. Warm but professional — like an executive assistant at a top firm.
The ProfessorAdaPatient, thorough, slightly academic. Loves explaining connections between concepts. Cites sources.
The PilotJetBrief, decisive, calm under pressure. No wasted words. Uses structured formats.
The CoachMayaWarm, encouraging, invested in your success. Asks reflective questions. Direct when needed.

When you switch presets, a confirmation dialog appears if the current SOUL.md content has been modified, since switching replaces the existing text. If the SOUL.md content matches the target preset exactly, the switch happens without confirmation.

A “Customized” badge appears next to the heading when the SOUL.md content no longer matches any preset — either because you edited it manually or started from scratch.

Click Customize below the preset picker to expand the SOUL.md editor. This is a Markdown file that defines the agent’s personality and identity. The agent reads it at the start of every conversation.

You can use a preset as a starting point and fine-tune the text, or write a completely custom personality from scratch. Editing the SOUL.md content after selecting a preset automatically marks the personality as “Customized.”

All presets include the instruction “Always respond in the same language the user writes in” — consider keeping this in custom personalities as well.

The Instructions tab provides an editor for the agent’s AGENTS.md file. This is separate from SOUL.md (personality) and serves a different purpose:

  • SOUL.md (Personality tab) — who the agent is: tone, style, character
  • AGENTS.md (Instructions tab) — what the agent does: tasks, domain-specific rules, operating procedures

Think of AGENTS.md as the agent’s job description. Use it to define what the agent should do, how it should handle specific tasks, and any domain-specific rules it should follow.

The agent reads both files at the start of every conversation. Separating personality from instructions makes it easy to change one without affecting the other — for example, switching an agent from “The Butler” to “The Pilot” personality without rewriting its task instructions.

Not sure whether a piece of knowledge belongs here or in Memory? See Instructions vs. Memory for the heuristic.

The Permissions tab controls which tools an agent can use. Pinchy uses an allow-list model: agents start with zero tools enabled, and admins grant specific capabilities.

Tools are organized into two categories:

Safe Tools — sandboxed access to admin-approved directories:

ToolDescription
List approved directoriesList files in directories the admin has selected
Read approved filesRead file contents from approved directories only

When a safe tool is enabled, the Allowed Directories picker appears so you can select which directories under /data/ the agent can access.

Powerful Tools — direct server access:

ToolDescription
Run commandsExecute shell commands on the server
Read any fileRead any file on the system, without restrictions
Write filesCreate and modify files on the system
Browse the webFetch web pages
Search the webPerform web searches

For a deeper explanation of the permission model, defense in depth, and how permissions are translated to OpenClaw configuration, see Agent Permissions.

The Access tab controls which users can see and use an agent. Without an enterprise license, this tab shows an upgrade prompt. See Groups for details on creating and managing groups.

Agent Settings — Access tab with visibility modes and group picker

ModeBehavior
All usersEvery authenticated user can see and chat with this agent.
RestrictedOnly admins and members of selected groups can access this agent.

New agents are created with Restricted visibility by default. This lets admins configure and test an agent before making it available.

When visibility is set to Restricted, a group picker appears listing all groups configured in Settings. Check one or more groups to grant their members access. If no groups are selected, only admins can see the agent.

Agent Settings — Telegram tab with bot connection status

The Telegram tab lets admins connect a Telegram bot directly to this agent, so users can chat with the agent from their phone. Each agent can have its own bot with its own Telegram username.

The main Pinchy bot must be set up first via Settings → Telegram — users link their Telegram accounts once via the main bot and can then message any bot they have permission to use. For Smithers, this tab shows the shared main bot status; to remove it, use “Remove Telegram for everyone” in Settings → Telegram. For other agents, the tab shows a token field and a Connect button.

For the full flow including BotFather setup and account linking, see Set Up Telegram.

Agents are created from the Create New Agent page. Pinchy ships with pre-configured templates for common use cases — Knowledge Base agents, vision-powered document analyzers, Odoo ERP assistants, email agents, and more. Each template auto-configures tools, personality, and system instructions; you choose a name and any template-specific options.

The generic starting points are:

  • Knowledge Base — pre-configured with safe file tools for answering questions from directories you select
  • Custom Agent — starts with no tools, for when you want to configure everything manually

Domain-specific templates (e.g., Sales Analyst, HR Analyst, Contract Analyzer) appear in the template picker when the corresponding integration is connected or their required capabilities are available. See Connect Odoo for the Odoo template list.

After creation, all settings can be changed at any time through the settings page described above. The runtime restarts automatically after creation to activate the new agent’s config.

Changes across all tabs are tracked independently. A small amber dot appears on any tab with unsaved changes, and the bottom bar shows “Unsaved changes.”

The save button behavior depends on what changed:

  • Save — appears when only personality, instructions, or access settings changed. Saves without disruption.
  • Save & Restart — appears when General (name, model) or Permissions settings changed. These changes require restarting the OpenClaw agent runtime to take effect. A confirmation dialog lists all pending changes and warns that active chats will be briefly disconnected.

If you try to navigate away with unsaved changes, a warning dialog asks you to confirm before leaving.