Nexevo developer docs

Send a task, the platform does the rest — it auto-routes to the right model, auto-calls the right tools, and streams the result back. No tool, model, or endpoint configuration needed.

Send a task — that's it

Send one field — {task} — to POST /v1/workspaces/{workspace_id}/agent/runs. The platform auto-routes to the best model, auto-calls the right tools, and streams the result. You never wire up tools, pick models, or choose endpoints.

bash

curl https://nexevo.ai/v1/workspaces/$NEXEVO_WORKSPACE/agent/runs \
  -H "Authorization: Bearer $NEXEVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"task": "Summarize today's top AI news."}'
# -> {"run_id": "…", "trace_id": "…"}

POST returns {run_id, trace_id}. Tail GET …/agent/runs/{run_id}/attach for the streamed result. See the Quickstart for the full 3-step flow.

What the platform handles for you

Auto-routing — the platform decides whether your task needs a quick chat reply or a full autonomous agent run. You send to one endpoint; it picks the lane.
Built-in tools — web search, your knowledge base, code execution, a real browser, file generation (PDF/docx/slides) — all auto-called as needed. You don't configure or select them. See How it works.
Streamed output — every step (planning, tool calls, reasoning, the final answer) streams back in real time. You just read the result.

Need more control? (all optional)

Every other parameter is an optional override — pin a model, set a budget_usd, attach a specialist role, or ground the run in a project's knowledge. None are required. See Agent runs for the full surface.

Ways to integrate

Method	Best for	Effort
`REST API`	Full control, any language	Send a task — see Quickstart
`OpenAI / Anthropic SDK`	Already using these SDKs	Change base URL + key — see Compatibility
`MCP server`	Claude Desktop, Cursor, MCP clients	Point at one URL — see MCP