Skip to content

Role Bundles

Instead of selecting individual scopes, you can provision an agent with a role bundle — a curated set of scopes plus a prompt template that tells your LLM what it should and shouldn’t do.

There is deliberately no “full access” role. An agent that needs everything is several agents — use separate agents with separate keys, each scoped to what it actually does.

Scopes: context:read · candidates:read · applications:write · jobs:read

Discovers and engages candidates before they apply. Can search the candidate pool, view rankings, express recruiter interest, and source candidates into the pipeline.

Best for: Automated outbound sourcing workflows, proactive talent discovery.


Scopes: context:read · applications:read · applications:write · jobs:read

Manages applications as they move through stages. Can list and update applications, view match explanations, and trigger ranking refreshes.

Best for: Workflow automation agents that triage the inbound queue, route candidates, and keep stages up to date.


Scopes: context:read · applications:read · messages:send · interviews:write

Handles candidate communication and scheduling after a match. Can read message threads, send messages, and schedule interviews.

Best for: Scheduling bots, follow-up automation, candidate experience automation post-match.


Scopes: context:read · analytics:read · jobs:read · applications:read

Read-only access to funnel data and pipeline state. Cannot take any actions.

Best for: Reporting agents, dashboards, data pipelines, weekly summary bots.


Scopes: context:read · jobs:read · jobs:write

Manages job postings only. Cannot see candidates or applications.

Best for: Agents that sync job postings from your ATS or HRIS, auto-create roles from approved requisitions.


When you create an API key with a role-bundled agent and omit explicit scopes, the key inherits the bundle’s scopes automatically. The server owns the mapping — you can’t accidentally grant more than the role allows.

You can still specify scopes manually to further restrict a key (e.g. give a pipeline agent only applications:read — no writes). You cannot grant scopes beyond the bundle’s own set.

When your key is tied to a role bundle, get_platform_context returns a role_directives field — a prompt-ready instruction block that tells your agent what to focus on and what to avoid. The portal also shows this template for copy-paste into your system prompt.

The directives from get_platform_context are the authoritative source. If you paste them into your system prompt at start-up, the agent’s behavior stays in sync with the role even if the bundle is updated.