Every surface of an agent that ships the fix
It acts, it knows, it ships, it connects. Attensira measures how often ChatGPT, Perplexity, Claude, Gemini and Google AI name and link you, then opens a pull request against the repo the losing page lives in. It never merges it.
Cancel anytime • Nothing goes live without you
It acts
The agent does the work, and every irreversible step stops for an explicit yes.
The agent
Ask in plain language, and it acts on the answer. Scoped to one workspace, limited to tools you grant, stopped for an explicit yes before anything irreversible.
Automations
Write an instruction once and the agent runs it on a cadence, with a full run history and an idempotency key so a retry never bills a second run.
Inbox
Every stop the agent makes, in one list. An unanswered approval expires after 7 days and the run closes having changed nothing.
Knowledge and skills
A brand profile and reusable named skills. It changes how the agent writes, never what it measures. Growth and Business.
Plugins
A granted plugin is the only way a tool enters the agent's toolset — and you can register your own MCP server so the agent can call it.
Credits
Agent work is metered; reading numbers you already collected is not. You pay for work done, not for seats or dashboards.
It knows
What assistants actually answer, measured as rates with the sample size attached.
Prompt tracking
A prompt is one buyer question, put to each tracked model on a schedule and answered for real. Each prompt runs at most once per model per country per day, and the verbatim answer is kept so the evidence is readable.
Mention rate
Successful runs naming your brand over all successful runs in the window. One definition, the sample size attached, and no change reported until it clears a 95% significance test.
Citation tracking
How often a model attached a link to your own domain, on the same denominator as mention rate. Mentioned is not cited, and the gap between the two is the signal.
Competitor tracking
Score rival domains over the answers you already collect. No extra runs, no prompt slots — and no share of voice that pretends to divide a fixed pie.
Topics
Group prompts into a buying question. Rollups are pooled rates, not averages of averages, so a topic proves a change sooner than the prompts inside it.
Model surfaces
Five surfaces with stable ids: chatgpt, perplexity, claude, gemini, google_ai. A new workspace starts on ChatGPT and Perplexity; Gemini unlocks on Growth, and Google AI has no API reader at all.
It ships
The output is a pull request, not a recommendation: the page rewritten, on a branch, for you to review.
It connects
Where Attensira reaches into the systems the pages already live in — your repositories, and your edge.
AI traffic and crawler logs
AI crawlers do not run JavaScript, so a browser beacon never sees one. Count their fetches in your edge middleware and find the paths they never reach.
GitHub
A first-party GitHub App, scoped to the repositories you select. The agent opens a pull request against the repo the page lives in — and never merges it.
MCP server
18 tools at mcp.attensira.com, reached from Claude, Claude Code, Cursor, VS Code, ChatGPT or n8n. Ten are read-only; no tool takes a workspace id.
REST API
The same 18 capabilities as plain HTTP under api.attensira.com/v1, with one workspace-scoped bearer key and no OAuth to complete.
Integrations
Every documented connection in one place — the GitHub App, MCP clients, your own MCP server, and edge middleware for crawler data.
Start with one question you want to win
Add the prompts your buyers ask. Attensira runs them against each tracked model, keeps the answer text, and shows the mention and citation rate with the number of runs behind it.
Cancel anytime • Nothing goes live without you