Build on Signal
Signal identifies the companies behind the anonymous traffic on your site. This is how you get that out of the dashboard and into whatever you actually work in.
Read
Accounts and visitors over a paginated JSON API, authenticated with a key you create yourself.
Subscribe
A POST to your endpoint the moment a visitor is identified, retried until it lands.
Ask
Natural-language questions answered from your own data, with the SQL it ran.
The public API is read-only. Keys reach reads. Changes are made in the dashboard, or by asking the copilot in a signed-in session, where each one is described and confirmed before it runs. If you need a write endpoint, tell us which one — that is how this list gets shorter.
For agents
Every page here is also served as its markdown source. Point an agent at these instead of scraping the HTML.
- /developers/llms.txt — the index
- /developers/llms-full.txt — every page, one fetch
- /developers/<page>.md — a single page's source
- /agent-api/capabilities — what this workspace's API can do, derived from the running service
Endpoint reference
Paths, parameters, error codes and exact limits live on geysera.com/developers, generated from the running service — so it cannot describe an endpoint that does not exist. This site is the part that explains how the product behaves.
Get started
From nothing to a first response.
Guides
How the product behaves, and what will surprise you.
Workflows
Whole jobs, end to end. Each one is a thing a customer actually asked for.
Receive identified visitors by webhook
Get a POST the moment a visitor is identified, and not miss any.
Export and keep in sync
Pull the account list once, then keep it current without refetching it all.
Ask questions in natural language
Use the copilot from code, and read the parts of the answer that matter.
Alert your team when a target account appears
Filter the webhook against your named-account list and post to Slack within minutes.
Build a daily call list
One call, a ranking you can read, and a prompt that is willing to say leave this one alone.
Write a first-touch email from what they read
An opener grounded in real behaviour, with the SKIP rule that stops it inventing one.
Spot a buying committee
Three people from one company in a week is one deal, not three leads.
Notice an account going quiet
The cheapest churn signal you have, weeks before the renewal conversation.
Fill your CRM with who is actually visiting
Compare the two account sets first; the gap is usually not the one you assumed.
Find the pages that create pipeline
Rank content by who it brings rather than how many, and read the coverage first.
Score and route inbound leads
Use what someone did before the form to decide who should get it.
Keep an ad audience in sync
Suppress customers, retarget high intent, and the consent questions to answer first.
A weekly revenue brief your LLM writes
Gather deterministically, write with a model, and propagate the disclosures.
Watch your own data quality
The workflow that protects every other workflow: knowing when the data went quiet.
Playbooks
Longer programmes, where the value is in the judgement between the steps.
Account-based marketing, end to end
Six stages, the two that carry the value, and the human gate to keep.
Turning content into pipeline
Measure content on who it brings, then prove the change with a causal test.
Let an LLM run your workspace
The response contract, refusals, the confirmation gate, and what a key can never do.
Case studies
What we found running this in production. Each one changed the product, and each is a mistake you can avoid making with your own data.
Most of your visitors were not people
69% of one account's events were unflagged crawlers.
The pipeline that reported success while doing nothing
A job that caught its own failure and completed, for days.
Why attribution covers less than you think
The ceiling is structural, and knowing the number is the useful part.