prompt → publish → call

One studio for prompts, agents, and what they cost.

Agent Studio is where a prompt becomes a published version, a version becomes an agent with tools, and every call lands in a cost report. Built for teams that run LLM workloads in production.

Sign-in required for every workspace.

agent run · text/event-streamlive

POST /projects/support-triage/versions/published/agent
data: {"delta":{"content":"Looking at the report…"}}
data: {"delta":{"toolCalls":[{"name":"Skill","args":{"name":"triage-rules"}}]}}
data: {"toolResult":{"name":"Skill","content":"# Triage rules…"}}
data: {"author":"escalation-agent","delta":{"content":"Severity: P2"}}
data: {"delta":{"content":"Filed as P2 with repro steps."}}
data: {"usage":{"inputTokens":812,"outputTokens":164,"costUsd":0.0031}}
data: [DONE]

projects

Projects & versions

Author prompts as versioned configs. Publish one version; callers pin it or follow the pointer.

agent

Agent loop

Multi-turn tool execution with turn budgets, skill loading, and transfer to sub-agents.

mcp

MCP tools

skills

Skills

Markdown behavior packs, listed to the model and loaded only when it asks.

chats

Chats

Talk to any agent project over streaming SSE, with tool results inline.

cost

Cost

Every call priced from the model registry and rolled up per project, per day.

Next.js · DynamoDB · one OpenAI-compatible channel for every model