Capri AI
Voice & TelephonyConnect AI agents to Capri AI, an AI receptionist that answers customer questions and books appointments. Agents read booked appointments and caller context and pick up the work that follows the call.
What This Integration Enables
Capri AI is not a phone system. It is the thing that answers, and what it says is entirely a function of the prompt and the knowledge sitting behind it. That is the part of a receptionist that decays quietly. The website gets updated, the holiday hours change, a service is retired, and the receptionist keeps confidently telling callers the old answer because nobody owns the job of re-teaching it.
FlowRunner agents take that job. Through this connector an agent can queue website pages into the receptionist's knowledge with Scrape Website, push question and answer pairs with Add FAQs, load a document from a URL with Upload Knowledge File, and stand up a whole new receptionist with Create Agent when a location or a product line opens. Query Agent is the part that matters most for confidence: it sends a message the way a caller would and returns not just the answer but every action the receptionist took during the turn, including the calendar it booked on, each with its resource name and timestamp. That makes the receptionist testable from a flow instead of testable by waiting for a complaint. What the agent will not do is decide on its own that new wording is good enough to say to every caller. That decision stops for a person, which is the human-in-the-loop default that runs through every FlowRunner connector.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Keeping the after-hours answer honest
A services business publishes updated pricing and revised opening hours. An agent watching the CMS picks up the change, runs Scrape Website on the affected URLs, and waits for the processing jobs to finish. It then runs Query Agent through a standing list of the twenty questions callers actually ask, including the awkward ones about cancellation and out-of-hours callouts. Two answers still quote the old price and one falls through to the fallback wording. The agent drafts a corrected prompt and a short set of FAQ entries, and puts both in front of the operations owner rather than publishing them. Once approved, Add Prompt and Add FAQs run together, and the agent re-runs the same twenty questions to confirm the fix.
Standing up a receptionist for a new location
A multi-site clinic opens a fourth location. The agent runs Create Agent seeded with the group's standard prompt and the location page URL, then Rename Agent so the record is identifiable in the Capri portal, then Upload Knowledge File to load the location's intake document from the group's document store. Scrape Website adds the location's service pages. Before the new number is pointed at it, the agent runs Query Agent through the group's standard question set so the new site launches with the same answers as the other three instead of whatever the model infers from a thin knowledge base.
Picking up the work the call started
Query Agent returns the intermediate actions the receptionist took during a turn, which is how a flow learns that a caller was booked onto a GoHighLevel calendar or written into a sheet. An agent reads that action list, matches the booking to the customer record, and creates the follow-up work: the prep task, the reminder, the intake form. When the action list shows a booking that collides with an existing appointment or lands outside the clinician's working pattern, the agent does not silently rebook. It raises the conflict as an automation exception for the scheduler to resolve.
Human-in-Loop Highlight
Add Prompt does not stage a draft. The prompt it adds becomes the active prompt, and from that moment it is what the receptionist says to every caller, including the ones who ring at two in the morning with nobody watching. So the agent treats prompt publication as the gate. It assembles the evidence first, the changed source pages, the proposed wording, and the side by side answers Query Agent produced under the old prompt and the new one, then posts to the owner of the phone line in Slack: "Proposed prompt change for the main line. Three answers change, two of them about cancellation fees. Old and new wording attached. Approve to make this the active prompt, or reply with edits." Nothing goes live until that reply comes back. The receptionist keeps answering with the prompt a person last approved.
Agent Capabilities
7 actionsConversations
1- Query Agent Sends a message to a Capri agent the way a caller would and returns the agent's final response along with every intermediate action it took during the turn, each with its resource name, action type and timestamp. Used to regression test a receptionist before a change goes live, and to discover what the receptionist actually did during a call so a flow can pick up the follow-up work. Passing a stable session ID keeps a multi-turn conversation together.
Agents
3- Create Agent Creates a new Capri agent and returns the full record including its generated ID, AI provider, model and failsafe provider. Used when a new location, product line or after-hours line needs its own receptionist, optionally seeded with a starting prompt and a website to learn from.
- Rename Agent Updates an agent's display name without touching its prompts, knowledge, actions or triggers. Used to keep agent records identifiable as the estate grows past a handful of lines.
- Add Prompt Adds a prompt to an agent and makes it the active prompt, returning the new prompt ID and the ID of the prompt that is now current. Used at the end of a review cycle, after a person has approved the wording. The custom fallback prompt controls what the receptionist says when it cannot answer from its knowledge.
Knowledge
3- Add FAQs Adds question and answer pairs to an agent's FAQ knowledge and returns the resulting list with a generated ID per entry. Used to correct a specific wrong answer precisely, without rewriting the whole prompt.
- Scrape Website Queues one or more URLs to be scraped into an agent's knowledge base and returns a processing job per URL. Used to keep the receptionist aligned with the published site whenever pricing, hours or service pages change.
- Upload Knowledge File Downloads a file from a publicly reachable URL and uploads it to an agent as a knowledge source, returning a processing job with the accepted MIME type and job ID. Used to load policy documents, intake forms and price sheets that never appear on the public website.
Frequently Asked Questions
What can FlowRunner do with Capri AI?
FlowRunner agents can run Query Agent, Create Agent, and Rename Agent in Capri AI, plus 4 more actions.
Does connecting Capri AI to FlowRunner require OAuth?
No. Capri AI connects to FlowRunner with an API key, no OAuth flow required.
Can Capri AI trigger a FlowRunner workflow automatically?
Capri AI doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Capri AI
$100 in credits. No card required. Connect in minutes.