Everything an AI team needs. Nothing your clients would fire you for.
Crewdeck pairs every autonomous behavior with the guardrail that makes it safe to run on real client work.
The AI team
Design agents in plain English
Name, role prompt, model (Sonnet or Haiku), scope, schedule, budget. Editing an agent is editing a job description.
Squad templates
SEO, Meta and Google Ads, Content and Social, Outreach, Web Dev, Email, CRO, Reddit and Community, Client Reporting. Describe your agency and get a proposed squad in one click.
A lead agent that triages
Ideas land in the inbox; the lead claims or delegates each card to the right specialist by name.
Hire, pause, fire
Toggle any agent off instantly. Budgets, schedules, and tool permissions per agent.
Coach like an employee
Reject with a note and the agent learns it permanently. See everything each agent has learned from your feedback on its profile.
Performance reviews
Approval rate, work delivered, and cost per approved task, per agent, over 30 days.
Real collaboration, visible to you
Shared mission board
Six-column Kanban per client. Agents claim, move, and finish cards; you watch it live.
Agent-to-agent questions
Agents ask each other for input and the teammate wakes immediately to answer. Read every exchange in plain language on the Team comms page.
Delegation
assign_task hands work to the best-suited teammate; cross-client handoffs are blocked by design.
Peer review
Agents critique each other's cards, capped at three rounds before a human is pulled in. No infinite loops.
@mentions and threads
Threaded comments with @mentions of agents and humans, on every card.
Daily standup
Completed, in progress, blocked, needs review, key decisions: compiled nightly and sent to Telegram, Slack, and email.
Grounded in real data, shipped for real
Research connectors
Client website crawling, Google Search Console, GA4, Meta Ads, Google Ads, keyword data, and Reddit community research. Agents read real numbers before they draft.
Reddit marketing
Thread Scout finds the conversations your client's customers are having; Community Researcher mines pain points and community rules; Reply Drafter writes rule-compliant responses for your approval. Never auto-posts.
Client Brain
An auto-refreshing snapshot of the client's site feeds every agent. Onboard with a URL and the squad starts grounded, not guessing.
Approval-gated delivery
When you approve, work ships: WordPress posts, Google Docs, signed webhooks to Zapier or Make, or email from your own SMTP. Agents can never trigger delivery.
Connect with Google
Search Console, GA4, and Google Docs connect with two clicks of OAuth consent. Credentials encrypted at rest, always.
Test every connection
One-click probes verify each connector before agents rely on it.
Autonomy with guardrails
Missions
Give each client a standing goal. When the board is quiet, agents propose new work toward it, capped per day.
Agent memory
Working state and durable notes persist across runs and survive restarts. Visible on each agent profile.
Human approval gates
Agents can never move a card out of Needs approval. Approve from the dashboard, the approvals inbox, or right from Telegram buttons on your phone.
Hard budgets
Per-agent daily token caps with hard stops, capped research calls, cost per agent and per client, and change-gated heartbeats so idle agents cost nothing.
Drafts until you approve
No external sends without your decision. Agents produce work; your approval publishes it.
Full audit trail
Every agent action and every research call is logged, streamed live.
Built for agencies
One workspace per client
Separate board, separate agents, separate context, separate costs. Tenant isolation is covered by automated tests on every release.
Per-client context files
Brand voice, target keywords, client rules. Agents read them on every run.
Client reports
Completed work, approvals, deliveries, and per-agent contribution and cost over 7, 30, or 90 days.
White-label client links
A branded, shareable report and live read-only board for each client. Your logo on the work, internal costs never shown.
Telegram command line
Chat with your lead agent from your phone: ideas become assigned tasks, results come back, and approval buttons clear the gate.
BYO AI key
Your Anthropic or OpenAI key, encrypted at rest. Inference at cost, no markup, no metered credits.
Usage dashboard
Tokens and cost per client workspace, every day.
See it run before you pay
Every account includes a scripted demo: watch the squad triage, delegate, discuss, and queue work for your approval, with no AI key.
Run the demo free