Loading…
Every hired agent reads the whole company's shared documents/memory, regardless of who uploaded them. Boundaries tell THIS agent what it may act on — it's an instruction, not an access restriction.
Reference knowledge specific to THIS agent — product details, FAQs, procedures — that its onboarding assistant draws on as data, not an instruction. The assistant interviews the owner using this text, so list what it should collect (e.g. menu, opening hours, brand voice). Type it directly, or upload a file to extract text into the box (review/edit before saving; the file itself isn't stored). Required for a new blueprint.
Business-specific creative direction added to this agent's daily poster prompt — e.g. for a restaurant: one real dish per poster, food photography style, never invent offers. Leave blank to use the generic poster prompt only. Applies to existing hires on their next poster run.
Shown on the pre-hire detail page. Anything left blank falls back to a placeholder.
When the agent is hired, Steward instantiates these across the billing cycle: a daily line creates one dated task per day; once creates a single task. Self tasks go to the agent, boss tasks to its reporting line. A condition can gate it (e.g. only create "provide documents" if the company-docs folder is empty).
No skills added yet — click + Add skill to pick from the catalog.
No skills in catalog — create some under Admin → Skills first.
Pick from the enabled channel catalog. Owners enable a subset at hire and choose approval mediums from them.
No enabled channels — add some under Admin → Channels first.
No connectors in catalog.
Only verified models (provider enabled + API key validated) are listed.
No verified LLMs — enable a provider and validate its key under Admin → LLMs first.
Protects margin against real LLM token cost, separate from the hours-based capacity tiers above. Set a target margin and Steward tracks actual token spend against a budget of monthly salary × (1 − margin) — e.g. a $100/mo hire at 30% margin pauses once real cost hits $70. Alerts fire at 75/90/100% of that budget.
monthly salary × (1 − margin)
No platform default is set — leaving this blank turns the guardrail off entirely for this blueprint. Set one →
Off by default. A company that hires this agent can further pause publishing on their own hire, but only once it's enabled here — they can't turn on what admin hasn't allowed.
Off by default. When on, the agent re-checks the company website daily; a change (or an owner request in the onboarding chat) becomes an update proposal the owner approves before anything is pushed to Facebook / Google, with ready-to-paste text for the website.
Off by default. Adds rating stats, a low-rating alert and an AI summary of what customers say. A source stays dormant (greyed out in the card) until its key or account is set up: Yelp needs YELP_API_KEY, Google needs GOOGLE_PLACES_API_KEY plus the company's Google Business link, Facebook needs a connected Page with the pages_read_user_content permission.