OrakulAn AI team for your agency — on your own server
SEO, reports, client support, contracts. Orakul automatically brings in the right agent for the task and delivers a finished file, not a plan for making one. Your data never leaves your perimeter.
After that — only tokens, ~$200/mo direct to the provider, no markup from us. Pays for itself in 4 months against one junior's salary.
Free, no signup. Orakul's agent checks whether ChatGPT, Gemini, Perplexity and other AI systems can see your site — and sends the report straight to chat. It's the same engine that runs inside agencies.
Orakul in plain English
Orakul is software that makes your agency's work dozens of times faster without sacrificing quality. It installs on your own server and builds you a team of AI employees: an SEO specialist, an analyst, a support manager, a legal assistant.
You describe a task in one sentence, in the web chat or on Telegram, and get a finished file back: a report, a proposal, a PDF, a website. A website takes 5–8 minutes. A second document for the same client arrives in seconds. The numbers come from Search Console, Google Analytics and PageSpeed. The answers come from your own policies and price lists. The model never fills gaps with guesses.
Put simply, Orakul is the strictest office manager you've ever had. It won't let your AI staff make things up: every number comes from your systems, every answer from your documents.
SEO audit for a prospect
Embed the audit on your website as a lead magnet. A visitor enters their site's URL, gets the audit three minutes later, and right below it sees a button that leads to you.
Commercial proposal
A finished PDF in your brand style. You set up the template once.
Weekly client report
The agent builds the report itself, on schedule. You just proofread it and send it to the client.
Junior hire or AI team: the two-year math
Total cost of ownership over two years: a salary plus payroll tax and a workspace, against a one-time setup fee and tokens you pay the provider directly. It counts what usually stays out of the estimate too: time to onboard, vacation and sick leave, and what it costs to scale from one role to five.
| Line item | Junior hire | Orakul "Team" |
|---|---|---|
| Start | $650 salary + 30% payroll tax | $2,000 one-time setup |
| Monthly | $850 + workspace | ~$200 tokens, direct to provider |
| Year 1 | ≈ $11,000 | $4,400 |
| Year 2 | ≈ $11,000 | $3,180 (tokens + $65/mo renewal) |
| Total, 2 years | $22,000 | $7,580 |
| Vacation, sick leave, turnover | yes | no |
| Time to onboard | 2–3 months | 1 business day |
| Scaling to 5 roles | 5× the cost | same tokens |
Savings over two years
The difference in total cost of ownership vs. one junior. With two hires, it's double.
Payback period
The monthly difference covers the setup cost by the fourth month of work.
To first result
20-minute deployment, process setup, and training — all in the same business day.
Calculated for a $650 base salary with 30% payroll tax and $800/year workspace cost. Token spend is real, based on our installations at 4–6 active roles.
Routine eats the day before you even sit down with clients
Scaling
Taking on two more clients means hiring someone. Hiring someone means –$11,000 a year and three months of onboarding.
Development
Hiring a developer for one bot costs more than the tool itself. And you need five bots.
Data
Client databases, contracts, and correspondence go off to someone else's cloud. Some clients simply won't accept that anymore.
Subscription zoo
Five services, five bills, five dashboards — and not one of them delivers a finished file.
For companies that sell time and expertise, not a product
Orakul was built for the work of an SEO agency — and what remains of that history is exactly the ready-made tools: PageSpeed, Search Console, analytics, backlink profile analysis. The platform itself isn't tied to an industry: roles, knowledge bases, scheduled work and finished documents are just as necessary anywhere people sell specialists' time — a digital agency, a web studio, a consultancy, a law or accounting firm. If the tools for your kind of work aren't on the list yet, we're ready to build them with you.
Digital & marketing agencies
Client reports, content plans, briefs, ad-account summaries, first-line support.
SEO agencies
Technical audits, semantic research, briefs for copywriters, rank and AI-answer visibility monitoring.
Web studios
Landing page copy, specs for developers, contracts and estimates, replies to inbound leads.
Roles — unlimited. Always.
We don't charge by the number of AI staff. Creating a new role takes three clicks in the panel: name, task, knowledge base. Thirty roles or three — the price doesn't change.
An orchestrator, not a chatbot
You set the task — the system picks the right role and model for it. No need to remember who to ask about what.
Four providers to choose from
OpenAI, Google, Anthropic, OpenRouter. The model is matched to the task: expensive for analysis, cheap for routine.
Delivers a file, not a plan
The output is a finished document, spreadsheet, or archive. Not "here's what needs doing" — done work.
Works in the background
The agent confirms the task and goes to work. The result lands in Telegram or the panel — no need to wait in chat.
One shared knowledge base
Your policies, price lists, contract templates, tone of voice. Agents answer as your company, not as a generic AI.
Cost control
The dashboard shows token spend by role and task. Limits are configurable — no surprises at month's end.
Agents work inside your Google Workspace
An employee connects their work Google account — and the agent sees their calendar, mail, Drive, spreadsheets and meeting transcripts. Not a shared company mailbox: access is per person, and the agent only reads the mail of whoever set the task.
Calendar and meetings
Checks availability, finds free slots and creates events — with a Google Meet link when one is needed.
Finds messages and reads the whole thread, then puts its reply in that same thread as a draft — you read it and send.
Drive and Docs
Searches and reads files, and saves its output as a finished Google Doc. The knowledge base can pull text from there too.
Spreadsheets
Reads data and appends rows: a report, a request log or a task list fills itself without manual copying.
Meeting notes
Finds the Meet transcript and turns it into minutes: attendees, decisions, action items with due dates.
Tasks and contacts
Files action items in Google Tasks and looks up someone's address by name when the request has no email in it.
.readonly suffix, so the token physically cannot change anything — it can't send an email, create a meeting or append a row. If a writing tool ever shipped by mistake, Google would refuse it; there's a test for that.Agency gets the rest: sending email, drafts and confirmation buttons, creating meetings and inviting attendees, Google Docs, appending spreadsheet rows, Meet transcripts, contacts, importing a Drive document into the knowledge base, a service account with domain-wide delegation, and the ready-made agents that ship with the product.
Nothing leaves without your click. The agent prepares the email or the meeting invite; a person sends it — "Send" and "Cancel" buttons appear in the web chat and arrive in Telegram.
This isn't distrust of the model. Emails and documents are written by outsiders, and an instruction hidden inside an email must never result in something being sent. The server acts on the saved text, so the content can't be swapped after you've seen it. Only the person the request belongs to can confirm it, and only within 24 hours. Scheduled tasks, which have no screen of their own, are confirmed the same way.
What your day looks like
Meetings in order with overlaps flagged, emails waiting on a reply, and what's worth knowing. On a schedule, before the day starts.
Inbox, sorted
Finds the messages you ask about, reads the thread and drafts a reply. If a person is named without an address, it looks them up in contacts.
From transcript to follow-up
Meet transcript → minutes → tasks with due dates → an email to attendees that goes out once you confirm it.
These three ship with the product and arrive with an update, including on installations that are already running. Your own agents are left alone — an update never overwrites a role you've edited.
Connecting takes a minute per person on the "My Google" page: the token is encrypted and stored against that employee, not in shared settings. For an agency there's a second route — a service account with domain-wide delegation, and then nobody has to connect anything by hand.
A Google Workspace subscription is required. Both routes work only with a corporate account on your own domain — a personal @gmail.com address won't do. The app is registered as Internal to your organisation, and domain-wide delegation is configured in the Google Workspace admin console: neither exists without the subscription.
From server access to your first working agent — one business day
Installed on your server
You can install it yourself by following the guide, or with our help — we install it, you watch. Either way it takes no special skills.
Tuning processes for you
We map out how work happens at your agency and build roles around real tasks, not abstract templates.
Loading the knowledge base
Policies, price lists, document templates, examples of good client replies.
Training your team
We show you how to assign tasks and create new roles yourselves. After that you don't depend on us.
What we need from you
A server running Ubuntu — 2 cores, 4 GB of memory and 40 GB of disk are enough (a CX23 at Hetzner, about €6 a month), SSH access, a domain with an A record pointing at that server, and your AI provider keys. We handle everything else. The full walkthrough with screenshots is in the installation guide.
What's included in $2,000
Setup takes 20 minutes — and we're not hiding that. You're not paying for the install; you're paying for agents that start working on your processes, with your documents and your tone with clients.
Included in the setup fee
- Deployed on your server — turnkey, environment included
- Source code handed over — immediately at deployment, full repo access
- 3 workflows configured for real tasks at your agency
- Unlimited roles — create as many as you need, never an extra charge
- Ready-made integration tools on every agent, no extra charge: GSC, Yandex Metrica, PageSpeed, Ahrefs (Rapid), Bitrix24, Google Workspace (Calendar, Gmail, Drive, Sheets, Meet, Tasks — with your own Workspace subscription), plus site parsing, Google search and an MCP browser-automation server (Playwright/Chromium)
- Knowledge base loaded: policies, price lists, document templates
- Telegram and web panel connected for the whole team
- Schedules configured — recurring tasks run themselves
- Bitrix24 integration
- Team training and instructions for creating new roles
- 12 months of updates and support, 4-hour SLA during business hours
Billed separately
No surprises here either — here's the full list of what you can pay extra for, and nothing hidden in the invoice.
| AI provider tokens | ~$200/mo direct to provider, no markup from us |
| Extra workflow beyond the package | $250 |
| Another business-system integration | $400–650 |
| Migration to another server | $250 |
| Renewal from month 13 | $65/mo or $650/yr |
We don't make money on tokens. The provider keys are yours, the bills come to you, spend is visible in the dashboard down to the cent. It's not in our interest for agents to "chat" more than necessary — that's a fundamental difference from services that sell marked-up internal "credits."
What we get compared to
Where we're stronger, and where other solutions do just as well. If you need a personal assistant for one person, get a cloud subscription — it's cheaper. Orakul is for when tasks come in a steady stream and data can't leave the building.
| Orakul | Claude Cowork | ChatGPT Business | n8n + API | Junior hire | |
|---|---|---|---|---|---|
| Where data lives | on your server | vendor cloud | vendor cloud | depends on setup | with you |
| Source code is yours | yes, immediately | no | no | yes, but it's your build | — |
| Choice of AI provider | 4 to choose from, switch per task | Anthropic only | OpenAI only | any, configured manually | — |
| Role limit | unlimited | within plan | billed per seat | unlimited | 1 person = 1 role |
| Needs a developer | no | no | no | yes, ongoing | no |
| Time to launch | 1 business day | 1 day | 1 day | 2–6 weeks | 2–3 months |
| Delivers a finished file | yes | yes | partially | you build it yourself | yes |
| Scheduled work | built in | limited | no | yes | yes |
| Who fixes what breaks | we do, 4-hour SLA | vendor support | vendor support | you do | the employee |
| Cost over 2 years | $7,580 | from ~$3,900 depends on seat count | from ~$3,250 + your time to set up | ~$5,200 + dev work and upkeep | $22,000 |
Comparison based on public vendor pricing and our own deployment experience. Cloud service costs are estimated for a 10-person team and depend on exchange rates and vendor pricing policy.
SEO agency, 14 people, Yekaterinburg
Before
Reports for 20 clients were compiled by hand — three full days at the end of every month. First-line support sat with account managers and ate up to a third of their working time. They didn't want to grow headcount for this.
Implemented
"Report Analyst" and "Support" roles, a knowledge base built from policies and price lists, a Google Search Console integration, results sent to Telegram. Launch — one business day.
Now
Reports for every client are compiled automatically on schedule. First-line support is handled by the agent; account managers only step in for complex requests. No hiring needed for this volume.
We expected this to take six months to roll out and a dedicated person to run it. In reality we launched in a day, and a month later we couldn't even remember how we used to compile reports by hand.
Head of Client Services
Why no company name
The client asked us not to disclose their name: Orakul gives them an edge over competitors in the same region, and they'd rather not advertise that. We respect that.
A reference call is available on request. If you're seriously considering deployment, we'll arrange a direct conversation with the client under NDA — they'll confirm the numbers and answer any question about how the system works.
What happens if it doesn't fit, or we disappear
Buying a $2,000 system for your own server is a decision that's scary to make. We answer all four questions upfront.
30-day pilot — $400
We deploy Orakul on our own servers and hand you the panel's superadmin password. It's the full Team plan — everything except access to the source code. We configure one process for a real task and — together, in writing — define the success criteria before we even start.
After a month you either continue — that $400 counts toward the full setup ($1,600 balance due), and everything built during the pilot (agents, roles, schedules) moves onto your own server as it stands, with nothing to set up again — or we refund it if the agreed result wasn't reached.
Continuity guarantee
Source code is handed over the moment we deploy — not when the company closes, not through escrow, not under special conditions. It's simply handed over together with the system.
That means: we can't cut you off, lock you out, or force a price hike on you. Orakul runs autonomously, with no calls back to our servers. The technology is patent-protected; terms of use for the code are set out in the license agreement.
Support and SLA
12 months of updates and support are included in the setup price. Response during business hours — up to 4 hours; critical incidents — up to 24 hours. Renewal after that costs $65/month — not a fee for new features: AI providers change their APIs several times a year, and we update the system so your agents keep working.
No lock-in
Stop renewing, and the system won't shut down or lock up. It keeps running on the version you have. The license for the version you bought is perpetual, the distribution and code are yours. You can walk away any time and lose nothing.
Roles — unlimited on every plan
We don't charge for the number of AI staff: creating a role takes three clicks, and billing for that would be strange. Plans differ by how many workflows we configure, whether market data is included, the extra capabilities, and support level.
- Roles: unlimited
- Up to 3 users
- The full Team plan, minus the source code
- Hosted on Orakul's servers — you hold the superadmin password
- 1 process we configure
- Success criteria fixed in writing before we start
- Refund if the result isn't reached
- Continue, and your agents, roles and schedules move to your own server
- Roles: unlimited
- Role creation rights — superadmin only
- Users: unlimited
- Baseline setup of your key workflows — up to 3
- Google Workspace: read-only — calendar, mail, Drive, Sheets, Tasks (
.readonlyscopes) - Source code immediately
- 12 months of updates, 4-hour SLA
- Team training
- Roles: unlimited
- Role creation rights — superadmin and users
- Users: unlimited
- Baseline setup of your key workflows — up to 5
- Google Workspace: read and create — email and meetings, documents, spreadsheet rows, Meet transcripts, contacts, domain service account, ready-made agents
- Source code immediately
- 12 months of updates, 2-hour SLA
- Dedicated support channel
- + Market data: Google & Yandex SERPs, search demand
- + Agent skills
- + Outcome tracking
- + Agent export/import
Renewal from month 13 — $65/mo or $650/yr on any plan: updates, provider API compatibility, support. You can skip renewal — the system keeps running. Tokens are billed directly to the provider, roughly $200/mo at 4–6 active roles. We don't make money on them.
What people ask before signing up
Do we get the source code?
Yes. Source code is handed over immediately at deployment — in full, no stripped-down builds, no closed modules. The technology is patent-protected, and terms of use for the code are set out in the license agreement: it prohibits resale and transfer to third parties, but doesn't restrict you from working with the system or customizing it.
How long does deployment take, and what do we need to provide?
Installing on the server — 20 minutes; a full launch with process setup, loading your policies, and team training — one business day. You'll need an Ubuntu 24.04 server (minimum 2 vCPU / 4 GB RAM / 40 GB SSD), SSH access, and AI provider keys. No server? We'll recommend a hosting configuration.
Why does setup cost $2,000 if installation takes 20 minutes?
You're not paying for the installation. You're paying for agents that start working on your actual processes: mapping out how work happens at your agency, building roles for real tasks, loading policies and templates into the knowledge base, setting up schedules and integrations, and training your team.
Why aren't roles limited, when competitors charge per "seat"?
Because limiting them makes no sense: creating a role is three clicks in the panel. Charging for clicks is a way to profit from friction, not from value. The more roles you create, the deeper the system embeds itself in your processes — that works in our favor, not against it.
What do tokens really cost?
On our installations — around $200/month at 4–6 active roles. The provider keys are yours, bills come to you directly, no markup from us. Spend by role and task is visible in the dashboard, and limits are configurable.
Does our data really never leave?
The system runs on your server: the database, files, correspondence history, and knowledge base stay inside your perimeter. The only outbound traffic is requests you send to your AI provider directly, through your own keys, bypassing us entirely.
What if it doesn't fit?
Start with a 30-day pilot for $400. Before we start, we agree in writing on what counts as success. Continue, and that $400 is credited toward the full setup. Didn't reach the agreed result — we refund it.
See how this works on your actual tasks
30 minutes: we go through your processes, show the system on real data, and work out whether deployment will pay off for you specifically. If not, we'll say so.
Or leave a request — we'll reply by email within one business day.