Ship production agents.Skip the plumbing.

Blazing Agents runs the hard infrastructure — sandboxed execution, async and scheduled tasks, skills and tools, and every agent resource as a managed API. You build the agent's logic; we run everything around it.

Durable Workspaces & Tasks

Multi-tenant from day one

No database to run

The platform, itemized.

Ten capabilities cover what production agents actually need. Configure them in the dashboard, reach them from the typed SDK — and know each one by its wire name.

ag_

Agents

Full configuration: Model, instructions, Skills, built-in Tools, MCP connections, Workspace attachment, lifecycle status, metadata, and avatar — one resource.
Built-in tool groups: Durable memory, structured task planning, and Workspace-backed file operations, shell execution, and Artifact publishing.
Immutable versions: Minted automatically on every update — compare history, pin runs to a version, restore any previous configuration.

Multi-tenant from day one.

Every resource lives inside a tenant, and first-class end-user attribution flows with it — through agents, sessions, memories, tasks, artifacts, and usage. One opaque userId, set at creation, scopes filtering and reporting everywhere.

one identifier, every surface

GET/v1/agents/ag_7Qk2f9Lm0Xr4Vh1/sessionsuserId=usr-4812

GET/v1/agents/ag_7Qk2f9Lm0Xr4Vh1/memoriesuserId=usr-4812

GET/v1/artifactsagentId=ag_7Qk2f9Lm0Xr4Vh1

GET/v1/usagegroupBy=user

Velocity is a feature.

A REST /v1 and a typed, resource-style TypeScript SDK share one set of contracts — runtime-validated, paginated, structured errors included. This is the entire wire: create an agent, chat with it, and hand the stream to your UI as a Vercel AI SDK response.

quickstart.ts@blazing-agents/sdk

import { BlazingAgents } from "@blazing-agents/sdk"; const ba = new BlazingAgents({ apiKey: process.env.BA_API_KEY });

new BlazingAgents()

One import, one key — seconds from an empty file to your own fully fledged agent platform.

const agent = await ba.agents. create({  name: "support-triage",  model: "claude-opus-4-8",});

ba.<resource>

Resource-style CRUD and query for agents, sessions, providers, prompts, tasks, artifacts, and memories — typed end to end.

// app/api/chat — the turn, straight into useChatexport async function POST(req: Request) {  const { message } = await req.json();  const chat = await ba.chat({ agentId: agent.id, message });  return chat.toResponse();}

toResponse() → useChat

The turn streams as UIMessageChunks on Vercel AI SDK v7 types — the response is byte-compatible with useChat, no shim.

agent skills

Your coding agent, fluent on day one.

Official agent skills are on the way — packaged integration knowledge that Claude Code, Cursor, Codex, or any skill-aware agent loads before it writes a line. Point yours at the skill and the integration builds itself.

blazing-agents/SKILL.md

---name: blazing-agentsdescription: Build on Blazing Agents —  SDK wiring, streaming chat, durable  Tasks, Workspaces, and Artifacts.---

$ba assist

Admin Agent · new session

›Delete the stale staging agent, then show this week's usage.

Tool approval requiredpending

deleteAgent { id: "ag_7Qk2f9Lm0Xr4Vh1" }

Approve? y/n

✓Agent deleted

✓Usage report ready · 142,608 tokens

resume: ba assist --session ss_4mX8pQ2vN7kL1cR6

One CLI, the whole control plane.

Chat with any hosted agent for rapid debugging — on your own machine or in CI — or hand platform operations to the Admin Agent, without leaving the terminal.

ba assist: Manage agents, tasks, providers, and usage conversationally. Sensitive changes stop for your explicit approval.
ba chat <agent>: Open a rich, resumable terminal conversation with any agent and its hosted tools.
ba run <agent>: Run one predictable turn from scripts or CI, with clean text, JSON, or schema-validated output.

Native credential storageDurable approvals