Analyze
Agents query your connected platforms live, surface anomalies, and explain what changed rather than handing you another dashboard to interpret.
No black boxes
Watch your marketing team plan, create, optimize, report
Specialist AI agents debate your strategy in a transcript you can read, disagree with each other in the open, and stop at an approval gate before anything leaves your workspace.
Free plan with 500 AI credits. No credit card, no sales call, and we delete your workspace and its data whenever you ask.
Scripted demo
Autumn retargeting — cost per acquisition review
Demo transcript
- Manager opened a discussion — autumn retargeting is under-delivering.
- Spend is up 18% week over week and cost per acquisition with it. Before anyone touches a budget I want the audience read and the creative read.
- Frequency on the 30-day retargeting pool reached 4.6, and the pool has not been refreshed since June. We are paying to re-reach people who already converted.
- Three of the four live ads run the same hook. I can ship two new angles from the brand kit without new photography.
- Then suppress converters for 30 days, cap frequency at 2.5, and move 20% of the retargeting budget to prospecting.
- Agreed on the exclusion and the cap. The budget shift goes to you as one reviewable change.
- Add a 30-day converter exclusion, set the frequency cap to 2.5, and shift 20% of the retargeting budget to prospecting.
- Your approval is needed — Nothing is published, spent, or changed until you decide.
- Approved — exclusion and frequency cap now, budget shift after Thursday's numbers.
- 13
- specialist agent roles, from audience analysis to LLM SEO
- 9
- action classes no autonomy level can auto-execute — including deletes, first-time public posts and any budget move over 20% a day
- 0
- silent actions — every run leaves a readable transcript
- 3
- interface languages, including full right-to-left Arabic
Usage counters appear here when the platform leaves private beta. We would rather show nothing than show numbers we cannot stand behind.
Built for how you actually work
The same agents, the same approval trail — a different shape depending on who is running the account.
Take on more clients without hiring ahead of revenue
Each client gets its own brand knowledge base, its own agents and its own approval trail inside one workspace, so the work on one account never bleeds into another.
Start free- Ten brands and 25 seats on the Agency plan
- Approval trails you can show a client verbatim
- Per-client brands, seats and approval trails
01 — 04
From connected to approved, in four steps
Setup is measured in minutes, and every step leaves something you can inspect afterwards.
Connect your platforms
Paste an access token for Meta, Google Ads, TikTok, LinkedIn or X. We call the platform's live API to validate it before saving, encrypt it at rest, and show you which account it resolved to.
Fill the brand knowledge base
Paste your positioning, claims, tone and offers. That knowledge is the only brand material the agents work from — you write it, so nothing is invented on your behalf.
The agents debate in the open
A Manager moderates the specialists through rounds you can read live — tool calls, confidence levels and disagreements included.
You approve, and the decision is on the record
Every publish, spend or platform change arrives as one reviewable proposal with the reasoning attached. Approving records your decision permanently and assembles the exact change; pushing it into the live ad account is still your click.
Four things the team does for you
Analysis, creative, publishing and optimization — the same loop a good marketing department runs, with the reasoning left visible.
Create
Copy, angles, video scripts and collateral written against your brand knowledge base — so the output sounds like you and only makes claims you approved.
Prepare to publish
Approved work is assembled into the exact change to make — copy, targeting and budget — with the approval record attached. One-click publishing straight into the connected account is the next thing we ship.
Optimize
Results come back into the discussion, so the next round argues from what actually happened instead of starting the conversation over.
Autonomy levels
Nothing ships without your yes
Autonomy is a ladder you set per brand, with a deny-list no level can buy its way past — not a switch you flip and hope.
- L1Default
Suggest only
Agents analyze, debate and propose; a person does the work. This is the default for every new workspace.
- L2
Draft without asking
Agents may run analysis, write internal drafts and generate images or video on their own. Anything that leaves your workspace still waits for your approval.
- L3
Approve by policy
Adds the reversible platform changes: pauses, budget decreases, negative keywords, creative rotation, bid and schedule adjustments. Anything outside your limits escalates back to you.
- L4
Spend within your ceiling
Adds budget increases, activating an approved campaign and republishing approved creative. The always-gated list — deletes, first public posts, billing and autonomy changes — still stops at you.
Every level writes the same permanent record: who decided, when, on what evidence, and with what comment.
On the roadmap
Dubinga MCP — your marketing brain, where you already work
Planned: connect Dubinga to Claude, ChatGPT, Cursor or n8n over the Model Context Protocol and query your own campaigns, transcripts and brand knowledge from there.
Planned clients
- Claude
- ChatGPT
- Cursor
- n8n
It is not built yet. When it ships it will be tenant-scoped and read-only, gated by the same organization permissions as the app — the free plan is the way to be there when it lands.
Where Dubinga sits
Data platforms move numbers and stop. Execution tools act but hide the reasoning. We are the layer that does both, out loud.
| Capability | Dubinga | Data & reporting platforms | Execution & creative tools |
|---|---|---|---|
| You can read the reasoning | Yes | Charts, no argument | Hidden agents |
| Turns findings into the exact change to make | Yes | Read-only pipelines | One channel each |
| Brand memory every agent works from | Yes | Not their job | Per-tool prompts |
| An approval gate you set, with a deny-list no level bypasses | Yes | Nothing to approve | Often automatic |
| Prices published, not gated behind a call | Yes | Often on request | Mixed |
| Bring your own model keys | Yes | Not applicable | Rare |
Categories, not named vendors: products move fast and we would rather you check today's feature list than trust a table we wrote.
Transparent pricing
Credits are the single usage unit, so a change in model prices never re-prices your plan. Every number below comes from the same catalog that bills you.
Billing period
Save 20%
Free
Free
- 500 AI credits per month
- 3 seats
- 1 brand
- 2 agent runs per month
Starter
$49 per month
- 5,000 AI credits per month
- 3 seats
- 1 brand
- 10 agent runs per month
Growth
Recommended$149 per month
- 25,000 AI credits per month
- 10 seats
- 3 brands
- 50 agent runs per month
- Image and video generation, 60s of video per month
Agency
$449 per month
- 100,000 AI credits per month
- 25 seats
- 10 brands
- Unlimited agent runs (fair use)
- Image and video generation
Enterprise
Talk to us
- Custom credit allowance
- Unlimited seats
- Unlimited brands
- Unlimited agent runs (fair use)
- Image and video generation
- Invoiced billing and custom DPA
Invoiced after a conversation: limits, data-processing terms and support are agreed with you.
Contact usEvery figure here is read from the same plan catalog that bills you. Credits, seats, brands and agent runs are the allowances your plan entitles you to; generated images and video run on your own fal or Replicate key.
Bring your own OpenAI-compatible key and your prompts run on your provider account, billed to you by the provider. Dubinga credits still meter the agent run itself.
Security posture, stated plainly
Tenant isolation in the database
Every tenant row is protected by PostgreSQL row-level security, enforced by the database rather than by application code remembering to filter.
Bring your own model keys
Store your own provider key and your prompts go straight to your provider account. Keys are encrypted at rest and never shown again after saving.
Approval audit trail
Every decision is stored with the actor, the timestamp, the evidence in the transcript and any comment — and it is not editable afterwards.
GDPR posture
Data-processing terms, export and deletion on request, and a documented subprocessor list. We do not sell data and we run no third-party ad trackers on this site.
SOC 2 — on the roadmap, not yet held
We are not SOC 2 certified today and will not imply otherwise. Type II is planned before enterprise general availability; ask us for the current status.
Questions people actually ask
If something here is unclear, that is our bug — write to hello@dubinga.com before you sign up.
Is the free plan really free?
Yes. The free plan needs no credit card, includes AI credits to run your first discussions, and does not expire into a trial. You upgrade when you need more credits.
What is a credit, exactly?
One abstract usage unit covering model tokens, generated images and generated video seconds. Pricing the abstraction rather than the provider means your plan does not change every time a model vendor changes rates.
Can I use my own model provider?
Yes. Save an OpenAI-compatible key in settings and your agents' requests go to your own provider account, billed to you by the provider. Dubinga credits still meter the run. Keys are encrypted at rest and never displayed again.
Will the agents spend money on their own?
Not unless you move them up the autonomy ladder yourself. Every new workspace starts at level 1, where agents only propose and you do the work. Even at the top level, deletes, first-time public posts, budget jumps over 20% a day, billing and autonomy changes still stop at you.
Who can see my data?
Your organization. Tenant rows are separated by PostgreSQL row-level security enforced in the database, and access inside your workspace follows your member roles. We do not train models on your data.
Can I get my data out?
Yes, by asking. Self-serve export is not built yet: write to privacy@dubinga.com and we send your campaigns, transcripts, approvals and knowledge-base contents as structured files, and delete the workspace if you want it gone. No lock-in clause, no export fee.
Which platforms do you connect to?
Meta, Google Ads, TikTok, LinkedIn, X and your own website are the first-class connections. You paste an access token, we validate it against the platform's live API before saving, and the connection reads your campaigns and metrics from there.
What if an agent is wrong?
That is what the transcript and the gate are for. You can read every round the agents argued, disagreement between them is left visible rather than smoothed over, and you can decline a proposal with a comment that goes back into the discussion.
Do you support languages other than English?
The interface ships in English, Spanish and Arabic, with full right-to-left layout for Arabic. Agents work in the language of your brand knowledge base and your brief.
See your first discussion in about ten minutes
Connect one platform, paste your brand basics, and read the team think.
Free plan with 500 AI credits. No credit card, no sales call, and we delete your workspace and its data whenever you ask.