How Obi works for you, from the first read of your company to the scan after a change ships.
Obi runs on the agent you already pay for.
Obi works inside Claude Code, Codex, Cursor or any terminal, on the power of your own agent. Your agent researches and reasons, with reads in parallel when it can. Obi's servers ask the AI engines: an answer your own agent writes never counts as a measurement.
Runs with: /obi
Every move starts from your buyers, in their words.
Before its first proposal, Obi reads everything you allow: your whole site, the reviews and alternatives a search returns, your rivals' home, product and pricing pages, and your demo videos, frame by frame. It proposes your ICP and its personas, and keeps what you say about yourself apart from what others say.
Runs with: obi business show
Obi reads the sales calls recorded in Granola: the questions and objections your buyers raise, in their own words. The buyer questions it scans are written from them, and from what buyers say on G2 and Reddit.
Reads
Granola
Tell Obi once: what to call you, a rival never to propose, the market to start with. It saves your words, applies them in every agent and every session, and forgets one when you ask.
Runs with: obi memory add --kind preference --text "answer in French"
Every draft is written in your voice: from the rules you gave, an example you shared, or the wording of your own pages. Before showing it, Obi swaps your name for a rival's. If the draft still holds true, it lacks your proof and goes back to the evidence.
Before writing anything new, Obi makes sure your pages can be used.
Obi reads your Search Console: the queries, the indexing, and the impressions in Google's AI features. Before it calls an absence from AI Overviews or AI Mode, it asks you for the one setting it cannot see, Search generative AI, and keeps your answer dated.
Reads
Google Search Console
Obi checks what AI crawlers meet on your site: robots.txt rules for GPTBot, ClaudeBot, PerplexityBot and the other named crawlers, noindex and canonical tags, your sitemap, llms.txt and firewall rules. Each finding names the engines its fix serves, and those whose own documentation says it is not needed. A robots.txt fix comes as a diff you review.
Runs with: obi readiness
Obi reads each key page twice: as the raw HTML a crawler fetches, and as a browser renders it. A price or a fact that appears only after JavaScript runs is flagged, since a reader of the raw HTML may miss it. Bing asks for critical content in the HTML; Google says it runs JavaScript, so Obi never sells this fix as one for Google.
Obi reports: Material fact appears only after rendering
Obi does the work, and reads the next scan to see what changed.
Obi looks for what the engines lack to recommend you to your buyer: first-hand evidence, or an explanation worth writing. It updates a useful page before it writes a competing one, and drafts the piece from your own data first.
Each priority is ranked by what it is worth to your buyer, the evidence behind it, its urgency, effort and speed, with its trade-off against the next best move. A fast fix can outrank a big content project. When the evidence supports nothing, Obi says so and leaves the list short.
Runs with: obi strategies
Obi opens pull requests on your product's repository for the pages and docs it writes. You review them and you merge: nothing goes live without you.
Reads
GitHub
When you tell Obi a change went live, it records the date and the publication, then compares the next scan on the questions that change targeted. A rise reads as a sign to confirm on later scans, never as proof the change caused it.
Runs with: obi harness outcomes --task-id <UUID>