Profiles & Personas

One agent,
many personas.

Profiles are named TOML overlays that shape how autonomous workers operate: prompts, model routing, depth and concurrency limits, denied tools, and other behavior controls. Use a built-in example or define the policy your workflow needs.

Built-in profiles

Built-in workflow examples

These presets ship with the runtime as starting points. Each adds a system prompt and selected parameter overrides; inspect or replace them before using in your own operation.

hunterBroad collection workflow

An example profile for collection-heavy work. It prioritizes extracting and enriching contact records, saves results through the configured tools, and allows parallel delegation within its profile limits.

analystRead-only analysis

An example profile for sourced analysis with cross-checking and no configured side effects. It denies save_contacts and git_push while the worker produces evidence and open questions.

validatorFocused verification

An example profile for checking and enriching an existing list. It uses verification tools and denies spawn_agent to keep this workflow in one focused worker loop.

Custom profiles

Your rules, your TOML

Create a TOML file in ~/.fathom/profiles/<name>.toml to override model, fast_model, temperature, max_depth, max_agents, max_iterations, timeout_seconds, replan_rounds and deny_tools. A user file with the same name takes priority over the built-in preset.

# ~/.fathom/profiles/my-profile.toml

model = "deepseek-chat"
fast_model = "deepseek-chat"
temperature = 0.3
# Set limits that fit your deployment
max_depth = 3
max_agents = 4
max_iterations = 30
timeout_seconds = 600
replan_rounds = 1
deny_tools = ["git_push"]

CLI usage

List, show, create, run

fathom profiles list

List all available profiles

fathom profiles show hunter

Show profile details

fathom profiles new my-profile

Create from template

fathom run --profile hunter "collect and verify records for Acme"

Run with a profile

Configure your workers

Use the personalization guide to define prompts, per-role models, tool deny lists and behavioral controls for your runtime.

Personalization docs