CloudRaker Run is live: the home for the agents your team will build
Enterprise agents, live in 30 seconds. Describe the job, or start from a ready-made agent. Run handles everything around it, from the phone line to the keys.
- Live in 30 seconds. Describe the job to Builder, or import one of about 100 ready-made agents for healthcare, insurance, sales, support and more.
- Where the work arrives. Give each agent a US or Canadian phone number, an email address and a place in Slack or Teams. Every call is recorded and transcribed.
- Agents that know your business. A knowledge base each agent reads folder by folder, and Structured Memory for invoices, cases and requests.
- Every run on the record. See what each agent did, every tool call, what it cost and whether it worked. Roll back any publish in one click.
- Agents as code. Each project is a git repo, a push deploys, branches get preview URLs, and secret keys never reach the model.
- Free up to 25 hours a month. Then $0.01 per agent minute, plus the model at its provider price. Early access opens in small batches.
Today we are opening CloudRaker Run to its first teams. Run is the home for all the agents your team will build: you describe the job, Run handles everything around it, and your team gets back to the work.
This note covers what we shipped, why it comes after everything else we've built, and how to get in.
Understand, decide, run
Run is the third step of a story we've been telling all year.
First, we taught machines to read and to listen. Paperwork parses documents and transcribes calls. Then we taught them to decide, in milliseconds: decision-machine-1 chooses the next step for each case. Run puts both to work. Agents take on the job, and they hand off to a teammate when a person must decide.
Most teams we talk to don't struggle with the model. They struggle with everything around it: the phone number, the inbox, the API keys, the tools the agent must reach, the documents it must know, and the question nobody can answer a month later, "what did that agent actually do?" Run is built for that part.
Describe the job, or start from a template
There are two ways to create an agent.
Builder. Tell Builder what the agent should do, in your own words: "triage new Linear issues", "answer support email from our docs", "summarize a PDF into a table". It writes the instructions and the settings, you run a test, and you publish when it works. Builder sessions are free.
Templates. Start from one of about 100 ready-made agents across healthcare, insurance, sales, support, finance, legal, operations and more. A prescription refill line. A first-notice-of-loss phone line. An insurance card reader. A lead qualifier. Import one, connect your tools, test it, and it's live.
Calls, emails and messages
An agent is only useful where the work arrives. In Run, each agent can get a phone number, an email address and a place in Slack or Teams, plus schedules, webhooks and an API. It answers on its own, and every conversation becomes a run you can review.
Phone is the one we're proudest of. Buy a US or Canadian number in the console and assign it to an agent. The agent answers, books, makes outbound calls and transfers the caller to a person when needed. Every call is recorded and transcribed into the run.
Teach agents your business
Agents need to know what your team knows. Run gives them two kinds of memory.
Knowledge. Write prices, policies and procedures as plain documents, in folders. They are indexed again on every publish, and each agent searches only the folders you allow. The pre-sales agent never reads support or HR files.
Structured Memory. Agents keep invoices, refill requests and support cases as typed records with dates, amounts, choices, emails and phone numbers. They create, find and update records during a run, and every value shows which run wrote it.
Tools connect once for the whole organization. Sign in to Linear, Stripe, HubSpot, Intercom or GitHub with OAuth, then choose which projects or agents can use each connection.
See every run
Every run is recorded, from the first message to the result: the conversation, every tool call, the duration, the credits and the model cost. The dashboard tracks runs, success rate, failed runs and cost over time, so you see when quality drops. Every publish is a version, and an earlier one goes back live in one click.
For developers: the same agents, as code
Your team builds in the console. Your developers get the same agents with git, an API and a command line.
- Agents live in git. Each project is a repository, and each folder is an agent with an
AGENTS.mdand anagent.yml. - A push deploys.
git pushchecks the files and deploys them. A branch gets its own preview URL. - An API for every agent. Start runs over HTTP and stream each step as events. Webhooks report the results.
- Keys stay out of the model. Agents write secret placeholders, and the real values go in on the way out.
Pay for the minutes your agents work
Run agents for free, up to 25 hours each month. After that, you pay $0.01 per agent minute, plus the model at its provider price, with no markup, or bring your own provider key.
A three-minute task costs $0.03. Phone calls cost $0.02 a minute (run time plus a phone fee), so a five-minute call costs $0.10, plus call routing at cost. A phone number is $2 a month. Indexing knowledge costs $1 per 1,000 chunks (unchanged files are free) and each search costs $0.02. Credits are shared with the Paperwork API.
Getting in
We're opening Run to teams in small batches. Request early access, create your organization and tell us the job you want an agent to do. We'll invite you when your spot is ready.
If you already have an invite code, sign up with it and put your first agent to work.