Chat

Send a list of messages and stream back the assistant's reply. The smart router picks a model for you (or pin one); group turns into a conversation for multi-turn context.

One call

Chat and Agent runs are two modes of the same system. Most of the time you don't need to distinguish — send a task to /agent/runs and the platform auto-routes. This page covers the /chat endpoint for when you specifically need a pure text reply (no tools, lowest latency, OpenAI-style messages array).

POST/v1/workspaces/{workspace_id}/chat —

send a list of messages and stream back the assistant's reply. Request body:

messages (required) — an array of {role, content}; role is system, user, assistant, or tool.
model — optional; omit (or leave empty) to let Nexevo's router pick the best model. Pin only to force a specific one.
max_tokens — defaults to 4096 (1–128000).
temperature — defaults to 0.7 (0–2).
intent_hint — a short task-type hint that bypasses the intent classifier when the caller already knows the kind of request.
conversation_id — thread this turn under an existing conversation. When set, the server rebuilds prior history from the database, so messages only needs to carry the new user turn.
project_id — run inside a project; its instructions and knowledge files are injected server-side (access-checked).
images — up to 4 image URLs (OSS) for vision chat; the router picks a vision-capable model and converts the last user turn into a multimodal message.

The reply streams as Server-Sent Events — one JSON object per data: line: {delta: "..."} for assistant text, {reasoning: "..."} for the chain-of-thought, an optional leading {context_compacted: {summary, turns_folded}} when earlier turns were summarized, and {error: "..."} on failure. The stream ends with data: [DONE]. (This is a different wire format from agent runs, whose events carry a type field.)

bash

curl -N https://nexevo.ai/v1/workspaces/$NEXEVO_WORKSPACE/chat \
  -H "Authorization: Bearer $NEXEVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{"role": "user", "content": "Explain SEPA Instant in two sentences."}]
  }'

Leave model empty to let Nexevo's router pick the best model for the task (quality / cost / latency balanced). Pin model only when you need a specific one.

Conversations (multi-turn)

To keep context across turns, create a conversation and append messages to it:

POST /v1/workspaces/{workspace_id}/conversations — create a thread (title, optional project_id).
POST /v1/workspaces/{workspace_id}/conversations/{conversation_id}/messages — append a message (role, content).
GET /v1/workspaces/{workspace_id}/conversations / GET /v1/workspaces/{workspace_id}/conversations/{conversation_id} — list / fetch threads.
PATCH /v1/workspaces/{workspace_id}/conversations/{conversation_id} — rename / archive.
POST /v1/workspaces/{workspace_id}/conversations/{conversation_id}/share / DELETE /v1/workspaces/{workspace_id}/conversations/{conversation_id}/share — mint / revoke a read-only public link.
DELETE /v1/workspaces/{workspace_id}/conversations/{conversation_id} — delete a thread.

Pass a conversation's project_id to ground the thread in a project's instructions + files.