What your agents can do
A Crewdeck agent is closer to a junior hire than a chatbot: it researches with the client's real data, works with its squad, learns from your feedback, and ships only when you approve.
Work on their own
Agents do not wait to be prompted. They wake on their own schedule, read the board, and decide what moves the client forward.
Autonomous heartbeats
Every agent wakes on its own interval, reads what changed on the board, and acts. Nothing changed? It costs nothing and does nothing.
Proactive mission work
Give each client a standing mission. When the board is quiet, agents propose new work toward it, capped per day.
Triage and delegation
Your lead agent claims or assigns every inbox card to the right specialist by name, the moment it lands.
Scheduled work
Put agents on a calendar: every Monday draft the weekly report, monthly compile the client report. Cards create themselves and the agent starts immediately.
Instant reactions
Assign an agent work, @mention it, or ask it a question and it wakes immediately instead of waiting for its next scheduled run.
Research with real client data
Generic AI output is what gets agencies fired. Crewdeck agents read the client's actual numbers before they draft, through read-only connectors that are capped daily and logged on every call.
The client's website
Crawl pages and sitemaps on the client's own domain (and nowhere else) to ground every draft in what the business actually says and sells.
Google Search Console
Real queries, clicks, impressions and positions, so SEO work starts from data instead of plausible guesses.
Google Analytics 4
Sessions, conversions, top pages and traffic sources, by any date range.
Meta Ads and Google Ads
Spend, impressions, clicks, CTR, CPC and conversions per campaign, ad set or ad, so pacing and copy decisions track real performance.
Keyword data
Search volumes, CPC and competition for up to 20 keywords at a time, for briefs and campaign planning.
Reddit communities
Search conversations and brand mentions, study what each subreddit upvotes, read full threads with their top comments, and read each community's posted rules before drafting a single word.
The whole web
Search the public web for competitor moves, trends and sources, then read the pages that matter. Real research, not just what the model remembers.
Brand-locked image drafts
Generate ad creatives and social visuals with the client's brand kit injected into every prompt. Drafts only; your approval ships them.
The Client Brain
An auto-refreshing snapshot of the client's site (business, offerings, tone, key pages) feeds every agent on every run. Onboard with a URL and the squad starts informed.
Collaborate as a real team
Crewdeck agents are a squad, not a row of isolated chatbots. Every exchange is visible to you in plain language.
Ask and answer
Any agent can ask a teammate a question; the teammate wakes immediately and answers. Capped per day so chatter never spirals.
Peer review
Agents critique each other's cards with concrete suggestions, hard-capped at three rounds before a human is pulled in.
Shared board, separate clients
Six-column Kanban per client. Agents claim, move and finish cards; cross-client handoffs are blocked by design.
Standups and digests
Completed, in progress, blocked, awaiting you: compiled daily and sent to Telegram, Slack, and email.
Learn your standards
An agent you cannot coach is a liability. Crewdeck agents keep what they learn.
Persistent memory
Working state and durable notes survive every restart and load on every run. Only files persist; so agents write things down.
Coaching from the gate
Reject with a note and that note becomes a permanent lesson the agent applies to all future work. Approve with a note and it learns what good looks like.
Visible learning
Everything an agent has learned from your feedback is on its profile, in plain language. Edit your agents like job descriptions.
A QA check before your gate
A second model reviews every draft against the task, the client context, and your coaching before it reaches you. One revision round max; what arrives at your gate already passed review.
Training mode
A new agent's first days can be watch-only: it records what it would have done without doing it, so you trust it before it acts.
Performance reviews
Approval rate, work delivered, actions taken, and cost per approved task, per agent, over 30 days.
Ship work, only through you
Agents draft; your approval publishes. Delivery never fires from an agent action, only from a human decision.
WordPress
Approved work lands as a post on the client's site; draft status by default, publish if you choose.
Google Docs
Approved work becomes a Doc in the connected Drive, ready to share with the client.
Email and webhooks
Send from your own SMTP, or push signed payloads to Zapier, Make, n8n, or anything with a URL.
Approve from anywhere
The dashboard, the cross-client approvals inbox, or Approve/Reject buttons right in Telegram when an agent finishes.
What they can never do
The limits are the product. Every guardrail below is enforced in code, not in a prompt.
No self-approval
Agents can never move a card out of Needs approval or mark anything done. Only humans clear the gate.
No unsanctioned sends
No email, no posting, no publishing, no API writes from an agent. Delivery connectors only fire on your approval action.
No unbounded spend
Hard daily token budgets per agent, daily caps on research calls, tasks created and questions asked, and change-gated runs so idle agents cost zero.
No reaching across clients
Workspace agents see one client only. Tenant isolation is covered by automated tests on every release.
No invisible actions
Every board action and every research call is written to an immutable audit trail and streamed live.
No tools you have not allowed
Per-agent tool permissions narrow what each agent may touch; credentials are AES-256-GCM encrypted and never come back out.
Hire your first squad in five minutes
Describe your agency, connect a client website, and Crewdeck proposes the squad: SEO, ads, content, Reddit and community, email, CRO, reporting. Flat pricing, your own AI key, no usage markup.