Agentic Control Center — the operating manual

Your daily mission control — what is urgent across every prospect and client, pulled live from the brain so nothing slips.

Every prospect in one ranked list — fit, credibility score, funnel stage, and the recommended next move, so you always know who to work next.

A deep page per prospect or client: contact, scan findings, grant opportunities, deliverables, and outreach — every fact sourced from that profile's brain file.

The GrantedSC Deep Scan and credibility score translate a nonprofit's public footprint into specific, fixable, fundable gaps — the substance behind every outreach.

Generate email and LinkedIn drafts in your voice, grounded in real specifics about the prospect. You review and send — nothing ever goes out on its own.

A single queue — with a live count in the nav — for everything waiting on your sign-off, chiefly drafted outreach. Preview it, approve and send, or mark it done; the badge clears as you go.

Your roster of AI teammates — see who does what, what each is cleared to know, and hire a new role-matched one in a click.

Talk to the whole team in one thread — pick who answers (@mention or toggles) or ask everyone at once. Dictate with your voice (mic), and a teammate can hand a task to the orchestrator in one click.

Everything every teammate has scheduled, on one surface — month, week, agenda, and a team swimlane — color-coded per agent and shown with their photo. Click any day or time slot to create and assign a task; agents can schedule for each other right from chat. Times run on Eastern (EDT).

Press ⌘K for a command palette that searches everything — views, agents, clients, products, tasks, settings — and jumps there instantly. Press ? for the keyboard-shortcut menu.

Drafted outreach lands in the review queue. Read it, edit the brain if needed, then approve and send from the app — or copy it into your own mail client. The funnel stage advances the moment it goes.

A dry-run renders the exact message that would go out — sender, recipient, subject, body — without sending anything, so you see precisely what lands before you commit.

Sending is one explicit click behind a confirm. A draft with an unfilled placeholder is refused, a warm-up cap limits how many leave per run, and a missing key fails safe — never a surprise blast.

A sent message is mirrored into your CRM as a note on the matching record, best-effort and automatic — so the CRM stays current without any copy-paste.

The next action for each prospect — first touch, follow up, book the call — is generated from its funnel stage as a task in your CRM, linked to the record and refreshed as the stage changes.

A single full-context agent routes work and assembles context for the rest. It is the only agent that ever touches the full brain or any sensitive detail.

Most teammates run scoped — sandboxed, with only the need-to-know slice the orchestrator hands them, never raw PII. Senior roles that can't work without the full picture (biz-ops; a direct-outreach agent that needs the contact list) can hold the full brain — and you grant or revoke that per agent behind an auth-gated toggle.

Each skill declares the data it touches and the tools it may use. Agents are provisioned with exactly the skills their role needs — nothing more.

Add a new employee with role-matched skills as data — no code change, and they're immediately chat-able in the same thread. New hires are always scoped; the platform can never mint a second full-brain agent.

It's a group chat. Tag one teammate, several, or everyone; each answers in their own voice. If a message isn't clearly theirs, they hand off or ask a quick clarifying question — like real coworkers, not a prompt box.

When a scoped teammate needs something only the orchestrator can pull, they flag it and you get a one-click hand-off. The orchestrator picks it up in the same thread and finishes the job.

Provision a code-capable engineer role and ask it for a change in plain language. It proposes the work on an isolated branch and hands back a diff for you to review — it never merges, pushes, or deploys on its own.

A graphic-designer agent generates brand-safe visuals — patterns, backgrounds, logos, social graphics — from a brief. For logos and headlines it composites real, crisp type over the AI art, so the spelling is never wrong.

Teammates answer in clean formatted text — headings, lists, bold — and one click copies a reply with its formatting intact, ready to paste straight into a doc or email.

Brain access is set per agent — scoped by default, full only where a role demands it — and changing it is gated behind a password. Scoped teammates physically cannot reach the full brain; the boundary is enforced by the workspace, not by trust.

The Agentic Control Center continuously scans for drift — a record created twice, a fact that disagrees with itself, an identifier shared across two orgs — and flags it to the orchestrator. So when an agent or teammate creates something where data already exists, you find out instead of quietly forking the truth.

Candid, off-the-record notes live in a private area of the brain that is excluded from the published data set. They never reach the hosted database or any client-facing screen.

Outreach is drafted against a sanitized copy of a profile with the private area stripped out, so an off-the-record note can never leak into an outbound message.

Every external action is drafted and staged for your review. Sending is one explicit click behind a confirm, capped during warm-up, and refused if a draft still has an unfilled placeholder. The platform waits; a human is always in the loop.

The reasoning runs on your machine against your own brain. The hosted app is a read-only view of a privacy-filtered snapshot — the full brain never leaves home.

When the engineer agent makes a change it happens on a throwaway branch in a sandboxed copy — never your live files. You review the diff and merge; nothing reaches the main branch, the remote, or a deploy without you.

Read any page aloud — or just the text you select, or each control as you focus or hover it. Free browser voices by default, or premium ElevenLabs voices when a key is set. The accessible sites the platform builds get the same reader, pre-recorded once so it plays free and offline.

Dictate into any chat with the mic. The browser's recognizer is free; ElevenLabs Scribe gives higher accuracy in every browser when a key is set.

High contrast, larger text, dyslexia-friendly font + spacing, a reading guide that tracks the cursor, a bigger cursor, stronger focus rings, always-underlined links, and a reduce-motion override — each one toggle, remembered per device.

Translate the interface — or a client site — into other languages on the fly (Gemini), and have the reader speak it in that language. English, Spanish, and more.

The hosted view sits behind a single access gate. Sign in once per device and the session is remembered — there are no per-screen logins. Locally, with no gate set, the app is wide open for you.

Product name, firm, region, audience, accent color, and logo come from a handful of environment variables. Or use the Brand Studio (/settings/brand) for a live-preview GUI that white-labels and provisions a fresh branded instance in one click.

One command scaffolds a fresh brain skeleton and brand file for a new client. It writes local files only and never takes an external action — so you can stand up a clean instance in seconds, then fill in the brain.