A runtime, not a chat box
Nexus gives your agents tools, budgets, traces and guardrails — the boring infrastructure that turns a clever demo into something you can actually put in front of customers.
Agents that plan, call tools, and settle
Describe the outcome, attach the tools, and the runtime handles the loop: plan, call, observe, adjust. Every step is recorded, so a run is never a black box you have to trust.
- Automatic tool selection with typed arguments
- Loop guards: max steps, max tokens, wall-clock budget
- Deterministic replay of any historical run
const run = await nexus.agents.run({
model: "nexus-1-pro",
input: "Refund order 8821 if it shipped late.",
tools: [lookupOrder, issueRefund],
max_steps: 6,
});
run.steps.forEach((s) => console.log(s.tool, s.latency_ms));Built for production, not for the demo
Autonomous Agents
Deploy agents that work 24/7 without supervision, with hard budgets and step limits you control.
Instant Processing
Real-time data analysis with sub-millisecond routing and a 241 ms median first token.
Enterprise Security
Bank-grade encryption for all your neural pathways. SOC 2 Type II, GDPR and CCPA compliant.
Custom Tools
Expose any HTTP endpoint as a typed tool. The agent decides when to call it — you decide what it may touch.
Run Traces
Replay any run step by step. See the arguments, the outputs, the latency, and the exact token bill.
First-class SDKs
TypeScript, Python and Go clients generated from the OpenAPI spec, so they never drift from the API.
Prompt Versioning
Promote prompts through staging like code. Roll back a bad revision without a deploy.
Model Routing
Route cheap tasks to Turbo and hard ones to Pro automatically, based on a complexity score.
See it run before you write any code
The playground uses the same models, the same temperature, the same token accounting.