Put agents to work in your product. None of the wiring.
You make the calls, and Novo runs them.
Ridgeline Help
My boots came in the wrong size. Can I swap them for a 10?
Looked up order 4821Done. A size 10 is on its way, and your return label is in your inbox.
Send a free exchange for order 4821?
A capable agent used to take all this.
- Model routing
- Keeping models current
- Prompt caching
- Context compaction
- Agent hosting
- Durable runs
- Resumable streams
- Code sandboxes
- Browser automation
- Web search
- MCP and API connections
- Office and PDF tooling
- Media generation
- Subagents
- Human approvals
- Prompt-injection defenses
- Secrets management
Now you just decide.
Give it a job. Novo runs the rest.
Set its context, its tools and the calls that need a yes. Novo runs the loop, the workspace and the retries, and streams each step back.
const novo = new NovoAgents();
const agent = await novo.agents.create({
context: 'Help Ridgeline customers with orders.',
capabilities: {
http: { connectionIds: [orders.id] },
},
approvals: { toolExecution: { hookId: review.id } },
});- Run started
- Called orders.lookup
- Asked for a yes
- Approved by your hook
- Reply streamed to your UI
Capable from the start.
Each capability is a field on the agent, with nothing to host. Big jobs split across subagents.
- Files and shell
- Office documents and PDFs
- A browser with saved sign-ins
- Web search with sources
- Images and video
- Speech, music and sound
- Subagents for big jobs
- Checks back on its own
- GitHub repos and pull requests
- Your MCP servers and APIs
Set the cost. Novo picks the model.
Novo tests models from the leading companies and serves the one that does each job best at your setting. A better model reaches your agent with no code change.
Novo picks the model for each capability at the Balanced setting.
Your product, your users’ yes.
Your transcript, files, UI and tool backends stay yours. Your approval hooks decide each eligible call, and your users give their yes in your own product.
Ridgeline Help
waiting for youWhich address should the new pair go to?
We never use customer content to train AI.
Serving stays within the disclosed US, EU and UK provider set.
History is stored encrypted.
For products whose core is something else.
The agent is a feature. Leave the stack behind it to us.
A support agent on your site
It looks things up through your API and asks your team before anything that needs a yes.
An assistant inside your app
It works on your users’ files, in your UI, with your product’s rules.
A back-office step
It works through records and documents and hands back the result.
Straight answers
When can I use it?
Developer access is closed for now while we focus on the helpers in Novo Work and Novo Life. Join the waitlist and we’ll write when it opens. There’s no date yet.
Which models does it use?
Each capability gets its own pick, checked on quality, speed and cost before it serves your work. You never pin a model or hold a provider key.
Do you train on my data?
No. We never use customer content to train AI. Serving stays within the disclosed US, EU and UK provider set, and history is stored encrypted.
How does pricing work?
You pay for usage, set a budget for each thread or run, and see usage by day, agent and thread, with each charge attributed.
What do I run on my side?
A server that calls one HTTP API through our TypeScript SDK. Your tools stay in your systems, reached through your MCP servers and OpenAPI connections.
Who is it for?
For developers who want an agent in their product without running the stack behind it.
Bring an agent into your product.
Developer access is closed for now. Join the waitlist to hear when it opens.
Building for your own team instead? See Novo Work