A buyer does the research in ChatGPT now, then asks the same chat to book the call or place the order. That last step is where most websites fall apart. Tobira founder Vlad wrote on Product Hunt that he asked his Claude to book a call with 10 real companies. The result, in his words, was zero bookings and zero payments. The agents hit account walls, booking pages that block AI, and forms built for fingers and eyes only.
Tobira Site Agent is his answer to that failure. It launched on Product Hunt this week as the missing piece of the Tobira network, which took the top Product of the Day spot in March 2026 and again with this launch, per the launch pages. The pitch is blunt. Your site should not wait to be read by an agent. It should send an agent of its own to meet it, answer it, and sell to it.
Tobira Site Agent
Official siteAn agent that lives on your website and sells to people and to their AI agents. Primary use case: turning AI traffic that already finds your business into booked calls, trial starts, leads and paid orders, instead of losing that traffic at a form the visiting agent cannot use.
Key features: a scan of your site that builds the knowledge base; answers drawn only from your pages, price lists, PDFs and FAQ entries; structured answers that visiting agents read in one pass; booking and payment inside the chat; voice conversations with a transcript; answers in the language of the visitor; a human handoff with context when a case gets hard; and a public address in the Tobira network so other agents can find you.
Advantages: the free route runs on your own key, so the model bill goes to your provider and not to Tobira; the rules are yours, in plain words, such as price ranges only or no promised discounts; private notes and internal documents stay private, per the maker; and you can manage the agent from Claude, ChatGPT, Codex or any MCP client, per the maker. Limitations: the whole idea needs buyers who actually send agents to shop, which is still a young habit; the paid tiers climb fast for a small business; and every speed and conversion figure published so far comes from the maker, not from an outside test.
Two visitors, one site
People see your usual website. Visiting agents get a version made for them, built by your agent from your own content. Tobira says the agent publishes the files agents look for, named llms.txt, agent.json and AGENTS.md, so a visiting agent can read the offer and talk to your agent directly instead of scraping page after page.
The token saving is the quiet part of the pitch. A normal pricing page can run to about 120,000 tokens of page weight, in the worked example on the Tobira site. The structured answer in that same example lands at about 1,300 tokens, which the company presents as 92 times fewer tokens for the agent to read. Treat that ratio as a company example, not a measured result for your site. The direction still makes sense. An agent that gets plan, price per seat, and the next free slot as clean data will finish the task faster than one that has to guess its way through your navigation.
The argument in one picture: pages built for eyes become data built for agents. Illustration: Dirty Static
That split is the same problem we covered in Sandbox Agent Sessions With Google Agent Substrate, from the other side of the counter. There, the question was how to run agents safely. Here, the question is what your shop window should do when one of those agents knocks.
What the agent actually does
The job list is sales work, not trivia. The agent asks about the case of the visitor, suggests the option that fits, books the call, and sends a payment link, right inside the chat the visitor started in. On the maker site, the sample outcomes run from a demo booked for a head of sales, to a quote request sent on as a lead, to a deposit held and paid by the agent of a client.
Three guardrails matter more than the tricks. First, the agent answers only from your materials. If the answer is not in your site, your price lists, your PDFs or your FAQ set, it offers a call instead of guessing, per the maker. Second, it shares only what you made public. Pricing and plans and security docs can be public, while internal notes and deal margins stay private. Third, a person steps in when it matters. Hard cases and open questions go to your team with the context of the chat attached.
Around that core sit the channels a small sales team already lives in. Leads and summaries land in WhatsApp, Telegram, email or a CRM such as HubSpot, Salesforce or Pipedrive, per the integration list on the maker site. The agent can also talk by voice, fill in security questionnaires from your security docs, and remember returning clients. If you want it to earn on its own, the Premium tier can sell subscriptions to an AI under your brand, built from your expertise, and Tobira keeps 15 percent of that subscription revenue, per the maker site.
Set up in minutes, on your own key
Tobira sells the setup as a five step path. The company scans your site and maps it, then writes the documents agents need. You add your own documents, set your tone and your rules in plain words, and test the agent on your real questions. It goes live on your site and for visiting agents, and leads start arriving in the channel you picked.
The founder puts the whole fix at 10 minutes on Product Hunt. The product page says live in 2 minutes. Both are maker claims, and the honest reading is minutes, not days, with the real work in checking what the scan learned before you trust it with prices. There is no code project to run here. You add a badge or widget to your site, and the agent is the product.
The network behind the shop window
A site agent alone is a better shop window. Tobira wants it to be a doorway into a market as well. Your agent gets a public address in the Tobira network, in the shape of an email address but for AI, such as the yourbusiness handle shown on the maker site.
When the agent of another company looks for what you offer, the two agents talk first and check the fit, per the maker. You get a qualified deal, or silence. The founder says the network holds more than 800 agents. That count is a company figure from the launch post, not an audited number, and the value of the network will stand or fall on how many real buyers and sellers join it.
What it costs
Start free with your own AI key. That is the headline on the pricing section of the maker site, and the launch offer on Product Hunt adds a free first week on the maker models, with the first month free if you add a card. After that, the paid plans buy you volume, voices, channels and help. Prices below are in USD, billed monthly and before taxes, checked on the maker site on November 7, 2026. Conversation counts are approximate, per the maker, since a short chat uses less than a long one.
Tobira paid plans, price per month in USD
List prices from tobira.ai, checked November 7, 2026. The free route with your own AI key costs $0 to Tobira, plus your own model bill.
| Plan | Price | Conversations a month | Replies with your own AI | Knowledge base | Team |
|---|---|---|---|---|---|
| Free | $0 with your own key | Set by your own key use | Set by your own key use | Built from your site scan | Not stated |
| Lite | $49 a month | About 300, about 700 agent replies | Up to 5,000 | About 1,000 pages | 1 |
| Start | $169 a month | About 900, about 2,200 agent replies | Up to 10,000 | About 2,000 pages | 1 |
| Business | $399 a month | About 2,400, about 6,000 agent replies | Up to 25,000 | About 5,000 pages | 3 |
| Premium | $799 a month | About 4,800, about 12,000 agent replies | Up to 50,000 | About 20,000 pages | 10 |
The steps between plans buy real things, not just higher caps. Start adds a choice of stronger models, Gemini among them, and a launch call with the founder. Business adds your own colours and fonts with no Tobira badge, client accounts, payments for consultations and services, voice mode up to about 1,500 minutes, and WhatsApp, Telegram and Instagram channels. Premium adds up to 3 sites, voice up to about 3,000 minutes, unlimited scenarios, weekly knowledge refreshes, and leads routed into Slack, Discord and Notion. Optional assisted setup runs from $899 on Start to $2,499 on Premium, and is free on some yearly billing routes, per the maker site. Every paid plan carries a 14 day money back guarantee, and the maker says a data processing agreement comes with every paid plan, with data handled under GDPR.
What Works
- Free route runs on your own key, with no Tobira fee
- Answers come only from your own materials, with a call offered when the answer is missing
- Booking, trial links and payment happen inside the visitor chat
- Your agent also gets a public address in a network of other agents
What Does Not
- Paid plans climb from $49 to $799 a month, a real bill for a small firm
- The speed claim of 2.2 times faster comes from tests by the founder, not an outside lab
- The whole model depends on buyers sending agents to shop, still a young habit
- Network size of more than 800 agents is a company figure, not an audited count
Which should you choose
Choose the free route if your buyers already arrive through AI and you lose them at the booking step. Connect your own key, let the scan build the agent, and test it with the exact questions your buyers ask ChatGPT about you. If the agent books one call your form would have lost, the idea has paid for its trial.
Choose Lite or Start if you sell calls, consultations or trials from your site. That is the shape Tobira was built for: SaaS demos, agency quotes, local services with a price list and a calendar. Start earns its jump from Lite when you want a stronger model on hard questions and more room each month.
Choose Business if the agent will take money and speak in your brand. Payments, client accounts, voice, and your own colours with no Tobira badge all land on this tier. For a firm that wants the agent to feel like staff, this is the first tier that reads that way.
Look elsewhere if your buyers still book like humans. If your traffic fills in forms, calls a number and pays on your site without trouble, a good booking page and a plain chat tool will serve you for less. Look elsewhere, too, if you cannot keep your price list and FAQ set current. An agent that answers only from your materials is only as good as those materials, and stale prices spoken with confidence are worse than no agent at all.
Final thoughts
The web was built in layers for people, then scraped by machines, and now it is being rebuilt for agents who act. Tobira is an early, concrete bet on that rebuild: one site for people, one clean version for agents, and a clerk who can close the deal in either room.
The tradeoffs are plain. You gain a door that agents can actually use, and you pay in plan fees, model bills, and the upkeep of the materials the agent speaks from. You also bet that your buyers will send agents at all. For a business whose website should sell and whose buyers already ask AI first, that bet looks worth a free test. For everyone else, watch the network count and the buyer habit, and keep your booking page in good shape in the meantime.
Facts checked November 7, 2026. This is a desk guide built from the maker site and the launch post, not a test on our own website. Speed figures, token figures and network size are maker claims and are labelled as such where they appear.
Sources
- Product, features, guardrails, integrations and pricing: Tobira, an AI agent that sells on your website
- Founder launch post, the 10 company booking test, the 2.2 times speed claim and the network count: Tobira.ai on Product Hunt
- Launch summary and pricing cross check: How to Use Tobira AI for Beginners, AI Tool Steps
- Hero and body images: generated illustrations by Dirty Static, in the house style. No maker screenshot is presented as the product interface.
END OF TRANSMISSION
Stay curious. Question everything.
