Agent Planners
A multi-agent harness that plans and runs marketing work across Google Ads, GA4, BigQuery, Meta and 250+ connected apps. It reads freely, but every write waits for your explicit approval before it ships.
What is Agent Planners?
Agent Planners is a multi-agent harness for marketing and ad operations, built on the opposite assumption from most marketing AI: an agent may read your accounts freely, but nothing changes until a human says yes. An orchestrator breaks a plain-language goal into steps and delegates them to six specialist agents, covering analytics, optimization, creative, research, safety and reporting. Each one runs on whatever model you pick through a built-in AI gateway, with GPT-5.5, Claude, Gemini, Qwen, DeepSeek, Llama and Mistral among the choices, switchable at any time.
Every request becomes an observable workflow. The agent clarifies anything ambiguous, judges what is needed, plans the steps and emits a typed plan you can read before it runs, showing its reasoning, the exact tool and the arguments it will pass. Reads run in the open; every mutation pauses at an approval gate, either in the app or through a one-time magic link sent by email. Rejecting one step skips that step instead of aborting the run, and every change lands in an audit log recording who acted, when, through which channel and with what before-and-after detail. Before a write executes, live account state is re-validated, so a plan built on stale data is rejected rather than applied blindly.
Natively, over OAuth, it connects Google Ads pinned to API v22 including MCC hierarchies, Display & Video 360, Campaign Manager 360, Google Ad Manager, AdSense, AdMob, Google Analytics 4, Search Console and BigQuery, alongside Meta Ads, TikTok Ads and Shopify. Composio opens 250+ further SaaS apps, and a custom MCP server exposes your own tools. A built-in toolbelt is selected automatically mid-run: web search, Google Trends, Firecrawl, Exa neural search, a hosted Chrome session, image and video generation, and a Python sandbox.
Per-account memory holds target CPA and ROAS, protected campaigns, excluded geos and pacing rules; successful runs are distilled into reusable Skills. Any goal can become a cron schedule in your own timezone. Organisations carry roles, isolated credentials and separate billing, and the product itself exposes a REST API and an MCP server so an external agent can drive it.
What it does
- Audit a Google Ads account across GA4, Search Console and BigQuery, then rank the next actions
- Pause campaigns below a ROAS target and shift budget towards the winners, with a before-and-after table
- Mine costly search terms with no conversions and propose campaign negative keyword lists
- Build paused Search and Performance Max campaigns, generating missing images and ad copy
- Join Google Ads cost with BigQuery revenue to compute profit-aware, true ROAS
- Track Search Console ranking drops and surface high-impression, low-CTR content gaps
- Turn a run into a client-ready PowerPoint deck, a written report or a Slack alert
When to use Agent Planners / When not to
A quick filter to help you decide if Agent Planners is the right fit.
When to use Agent Planners
- Paid media specialists and agencies managing several Google Ads accounts under one MCC
- Marketing teams wary of letting AI touch live budgets without a human gate on every change
- Data-literate teams whose real revenue sits in BigQuery or a CRM and who want true ROAS
- Publishers monetising through AdSense, AdMob and Google Ad Manager alongside their ad spend
- Lean teams and freelancers who repeat the same weekly reviews and want them on a schedule
When not to use Agent Planners
- Anyone who needs a mobile app: there is no iOS or Android build, only the web workspace
- Developers expecting a documented API reference, since the REST API is announced but never documented
- EU data controllers who require a Data Processing Agreement, an Article 27 representative or SOC 2
- Buyers who need a named, registered vendor with a postal address and a stated governing jurisdiction
- Anyone hoping to run more than one or two tasks a month without paying, given the free credit allowance
How to use Agent Planners
A typical end-to-end flow, from setup to results.
- Sign up on the website: nothing to install and no credit card required
- Create your organisation and set each member's role as admin, editor or member
- Connect a platform on the Integrations page through a standard OAuth flow
- Repeat for the other native platforms you need, from GA4 and Search Console to BigQuery and DV360
- Optionally add a Composio key or register a custom MCP server to widen the toolbelt
- Select the accounts a task should run against in the context picker, one or many at once
- Write your goal in plain language in the Workspace
- Answer the agent's clarifying questions about accounts, thresholds and the real objective
- Read the generated plan, then approve each write in the app or from the magic-link email
- Turn the goal into a cron schedule with an approver email if it should repeat
Pros & Cons
Pros
- Nothing changes on a live account without an explicit human approval
- The plan is readable before it runs, with the reasoning, the exact tool and its arguments
- A complete audit log, plus live re-validation of account state before every write
- Unusually wide Google coverage: acquisition and publisher monetisation handled by the same agent
- BigQuery joins let pause and budget decisions run on revenue figures you actually believe
- Every feature ships on every plan, free tier included, since plans differ only by credits
- No per-seat billing, and Bring Your Own Model removes the largest cost line for anyone already paying for a model
Cons
- The vendor is unidentifiable: no registered company name, no postal address, no stated jurisdiction
- A single mailbox handles support, legal questions and GDPR requests alike
- No Data Processing Agreement, no Article 27 representative, no named DPO and no security certification
- The domain was registered in May 2026 with no archived history, so there is no track record to check
- A REST API is advertised in the FAQ, yet no public API documentation exists anywhere on the site
- No mobile application, and plan credits reset each period without rolling over
- The free plan covers roughly one or two tasks a month, and no enterprise pricing is published
Pricing & Plans
There is a permanent free plan of 2,500 credits per month, with no credit card required, enough for roughly one or two tasks. The cheapest paid tier is Lite at USD 29.90 per month, falling to USD 23 per month when billed annually. Billing is per organisation rather than per seat, and usage is metered in credits, a credit being a fixed slice of real model and tool cost, around USD 0.001 on the Starter tier.
- USD 0 per month
- 2
- 500 credits
- no credit card
- roughly 1 to 2 tasks per month
- USD 29.90 per month
- or USD 23 per month billed annually
- 30
- 000 credits
- roughly 21 to 30 tasks
- USD 100 per month
- 100
- 000 credits
- roughly 71 to 100 tasks
- flagged as the popular tier
- USD 500 per month
- 550
- 000 credits including a 10% volume bonus
- roughly 390 to 550 tasks
- USD 1
- 000 per month
- 1
- 200
- 000 credits including a 20% volume bonus
- roughly 850 to 1
- 200 tasks
- custom volume
- arranged by email with the vendor
Data, GDPR & hosting
A consolidated view of how Agent Planners handles your data.
GDPR overview
Implementation is partial and unevenly documented. The privacy policy, effective 29 May 2026, grants data-subject rights explicitly, naming access, correction, portability, restriction, deletion and objection, and citing the GDPR for EU residents and the CCPA for Californians. Requests go to a single address and are answered within the legally mandated timeframe; account deletion is actioned within 30 days. Sub-processors are named individually, credentials are encrypted at rest and the service is closed to under-16s. What is missing is equally concrete: the site never states that it is GDPR compliant, publishes no Data Processing Agreement, designates no Article 27 representative and names no Data Protection Officer. Transfers to the United States rest on user consent, with no mention of standard contractual clauses.
Who owns the data?
The terms are unambiguous: your content stays yours. Goals, plans, custom Skills, reports and uploaded assets remain your property, and the vendor takes only a worldwide licence to process that content for the sole purpose of running the service. The platform itself, its code, branding and logos, remains the vendor's intellectual property. One exception deserves attention: any Skill you flag for cross-organisation sharing becomes readable by other organisations on the platform, which the terms treat as your explicit consent to share it, and which you can switch off at any time. OAuth credentials are encrypted at rest, scoped to a single organisation and never returned to the browser.
Reuse rights
Nothing in the terms restricts what you do with what you produce. Plans, reports, decks, ad copy and analyses are yours to reuse, publish or hand to a client without asking permission: the licence you grant runs one way and is limited to operating the service. On the vendor's side, the privacy policy states that data reached through the Google API scopes serves only user-facing features, with no advertising use, no resale, no training of generalised AI models, and no human reading it except with your consent, for security reasons, to comply with the law, or in aggregated form. Prompts carry your goal plus context drawn from connected accounts, and responses are stored against the task they belong to.
Data retention & training
Hosting summary
Everything runs in the United States, and the vendor says so plainly: the privacy policy asks users to consent to their information being transferred to and processed in the United States and in whatever other countries its service providers operate from. No European region is offered, and no choice of hosting region exists. The sub-processor list is named rather than generic, which is more than many vendors publish. Hosting and the workflow runtime run on Vercel, the database on MongoDB Atlas, file storage on Vercel Blob and authentication on Clerk. Model calls pass through the Vercel AI Gateway, which routes to providers including Anthropic, Google and OpenAI. The Python sandbox is Vercel Sandbox, transactional email goes through Resend, and web lookups use Perplexity and searchapi.io. The domain itself resolves to a United States address on Amazon infrastructure. OAuth credentials and connector keys are encrypted at rest, scoped to a single organisation and never returned to the browser. What is absent is contractual rather than technical: no Data Processing Agreement, no hosting certification and no standard contractual clauses covering the transfer are mentioned anywhere.
Things to keep in mind
Risks and trade-offs to weigh before adopting Agent Planners.
- The agent holds OAuth write scopes on live ad accounts, so the approval gate is the only barrier between a generated plan and real spend
- Approval fatigue is the practical risk: clicking yes on the twentieth plan of the week is how a bad change slips through
- The vendor states plainly that plans and reports may be incomplete, biased or wrong, and the terms make you solely responsible for whatever you approve
- Autonomous mode and pre-approval rules remove the gate deliberately, and enabling them transfers the risk squarely back to you
- Leaning on the agent for account judgement can erode a practitioner's own feel for a campaign over time
- You cannot identify or locate the vendor, which limits legal recourse and complicates any European compliance review
- Skills flagged for cross-organisation sharing become readable by other customers, so check what your methodology reveals before switching it on
Setup & Integrations
Technical difficulty
Low for the intended user. Nothing is installed: you sign up, create an organisation, click Connect on a platform card, authorise on Google's consent screen and write a goal in plain language. The vendor describes the path from sign-up to first optimisation as a five-minute one. The real prerequisite is access rights: you must own or administer the accounts you connect. Two optional steps raise the bar, since Shopify needs an Admin token with chosen scopes and a custom MCP server means hosting a Streamable-HTTP endpoint. Domain fluency in GAQL, ROAS and MCC structure matters more than technical skill.
Deployment
Integrations
Supported languages
Behind Agent Planners
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Agent Planners.
Frequently asked questions
Can the agent change my ad account without asking me?
Which platforms does it connect to natively?
Can I choose which AI model each agent runs on?
How do credits work, and what does one task cost?
Do unused credits carry over to the next month?
Can I point the agents at my own model deployment?
Is my data used to train AI models?
Who is behind the service, and where is it established?
Can I drive it from Claude, Codex or my own scripts?
What is the minimum age to use it?
Should you pick Agent Planners?
Agent Planners makes one architectural choice and builds everything around it: an AI agent may read your accounts freely, but it cannot change anything until you say yes. That single gate, with the readable plan, the audit log and the live re-validation of account state behind it, is what separates it from tools that ask you to flip a switch and hope for the best.
What surrounds the gate is unusually complete for such a young product. Nine Google platforms connect natively, covering both acquisition and publisher monetisation, with Meta Ads, TikTok Ads and Shopify alongside them. Composio and custom MCP servers extend the reach to hundreds of apps, while BigQuery joins let budget decisions run on revenue you actually trust rather than in-platform conversion estimates. Per-account memory, auto-built Skills and cron schedules stop the routine work from staying routine.
The reservations concern the vendor, not the product. Nobody is named. There is no registered company, no postal address, no jurisdiction, no security certification, no Data Processing Agreement and no Article 27 representative, and a single mailbox answers support, legal and privacy requests alike. The domain was registered in May 2026 with no archived history, so there is nothing to verify against. Handing OAuth write scopes on live ad accounts to a vendor you cannot identify is a real decision, and the approval gate mitigates that risk without erasing it.
The sensible path is the one the pricing already suggests. Start on the free tier, where every feature is present and only the credit allowance is small, run a couple of read-only audits, and judge the quality of the plans it produces before connecting anything that writes.
- Choosing a selection results in a full page refresh.
- Opens in a new window.