Agents & Personalities

Meowbert gives you two independent knobs to shape how the AI agent behaves: Agents control the model and reasoning effort, and Personalities control the tone and communication style.

---

Agents

An agent preset defines which model runs the task and how much reasoning effort it applies. You select an agent in the Task Composer before submitting a task.

Built-in agent presets

Preset	Reasoning effort	Best for
Default	High	Most tasks — balanced quality and speed. This is the safe choice when unsure.
Deep Think	Extra-high (xhigh)	Hard problems: complex architectures, multi-step analyses, subtle bugs. Slower but more thorough.
Fast	Low	Quick tasks: simple transformations, short summaries, fast iterations. Highest throughput.

How to select an agent

1. Open the Task Composer (click New Task in the sidebar).
2. Look for the Agent selector — it appears below or alongside the prompt input.
3. Click it to open the dropdown and choose a preset.
4. The selected preset is remembered as a draft for this project, so you don't need to re-select it each time.

If no agent is explicitly selected, the Default preset is used.

Agent presets and the model

Each preset carries a model identifier and reasoning parameters in its payload. By default all built-in presets use the same underlying model but differ in reasoning effort. Platform admins can configure custom presets with different models entirely — for example, a cost-optimized preset on a smaller model or a premium preset on a frontier model.

Admin-level customization

Platform administrators can define custom agent presets from the Admin → Agent Presets panel. Each preset requires:

• ID — unique lowercase identifier (used internally and saved in drafts).
• Name — display name shown in the selector.
• Description — short explanation of when to use this preset.
• requiresSuperAdmin — optional boolean; when true, only super-admin accounts can see and select the preset.
• Payload — JSON object specifying at minimum model plus a nested responses object for provider-facing request settings such as reasoning.effort.

Example custom preset payload:

[
  {
    "id": "premium-review",
    "name": "Premium review",
    "description": "Stronger model for critical reviews.",
    "requiresSuperAdmin": true,
    "payload": {
      "model": "gpt-4.1",
      "responses": {
        "reasoning": {
          "effort": "medium"
        }
      }
    }
  }
]

Changes to presets take effect immediately for new tasks. Existing running tasks continue using the preset they were started with.

---

Personalities

A personality is a tone and communication style profile applied to every task in a project. While agents affect how much the model thinks, personalities affect how it speaks.

Available personalities

ID	Label	Description
default	Default	Professional and neutral. Rejects ideological framing for technical tasks. Good general-purpose tone.
donald-trump	Donald Trump	Trump-style rhetoric: punchy, combative, high-confidence, America First framing.
none	None	No personality injection — the model uses its own defaults entirely.
concise	Concise	Fewest words possible. Short bullets, compact structure, no extra context unless asked.
friendly	Friendly	Warm, conversational, encouraging. Includes helpful transitions and extra practical tips.
cold	Cold	Emotionally neutral and clinical. Precise, impersonal, fact-first. No small talk.
aggressive	Aggressive	Blunt and confrontational. Challenges weak reasoning directly. Low hedge.
liberal	Liberal	Progressive framing with an argumentative style. For role-play or contrast testing.

How to set a personality for a project

1. Open the project in the sidebar.
2. Click Settings.
3. Find the Personality dropdown.
4. Select the desired personality.
5. Save.

The personality takes effect on the next task run in that project. It has no effect on tasks that are already running.

How personalities work

When a task starts, the worker reads the project's personality setting and injects the corresponding prompt text at the beginning of the agent's system prompt. For example, choosing Concise prepends:

"Be concise and direct. Use the fewest words needed to deliver a complete and correct answer. Prefer short bullets and compact structure over long prose."

Choosing None injects nothing, letting the model speak in its own default voice.

Choosing the right personality

• For automated pipelines (connectors, scheduled tasks): Concise or Cold reduce response bloat.
• For interactive use with non-technical team members: Friendly makes outputs easier to read.
• For thorough technical analysis: Default or Cold keep responses focused.
• For experimentation or testing model behavior: Aggressive, Liberal, or Donald Trump are useful contrast cases.

Custom personalities

Personalities are defined as .md files in the prompt-texts/personality/ directory. Each file name (without the extension) becomes a personality ID. To add a custom personality:

1. Create a new file, e.g., prompt-texts/personality/terse.md.
2. Write the instruction text that should be injected into the system prompt.
3. Restart the API and worker processes for the new personality to be loaded.

The personality will then appear in the Project Settings dropdown automatically.