Introducing Agentblit

We're excited to introduce Agentblit — a workspace for configuring, connecting, running, and inspecting production AI agents.

Why we built Agentblit

Most teams can prototype an agent in an afternoon. Shipping one is a different problem. Model configuration ends up in application code, every integration needs its own authentication flow, sensitive actions need approval, and a failed run is difficult to reconstruct after the fact.

Agentblit brings those operational pieces together:

• Workspace configuration — manage an agent's model, system prompt, tools, sub-agents, chat appearance, and API keys in the console
• Connected tools — attach MCP servers, HTTP connectors, and built-in integrations without putting connector credentials in your agent code
• Human control — set each tool to Allow, Ask first, or Block, and collect structured user input during a run
• Memory — summarize recent conversation history and give agents access to workspace memory that persists across sessions
• Runtime features — stream responses, accept multimodal input, and complete multi-round tool workflows
• Events — inspect each session as a timeline of prompts, model calls, tool calls, errors, token usage, and latency
• Multiple model providers — route requests to OpenAI, Anthropic, Gemini, or OpenRouter

Use the console or bring the SDK

You can run an agent through Agentblit's hosted chat, embed its chat UI, invoke it on a schedule, or use the Python and TypeScript SDKs inside your own application. Agentblit can also build and deploy applications from Git repositories and provision Postgres, Redis, or MySQL alongside them.

The SDK loads the model, system prompt, agent ID, and remote tools from Agentblit on the first run. Local tools can stay in your application and be registered alongside connected tools.

import { Agent, tool } from "agentblit";

const getDeploymentStatus = tool({
  name: "get_deployment_status",
  description: "Get the status of a deployment.",
  inputSchema: {
    type: "object",
    properties: {
      deploymentId: { type: "string" },
    },
    required: ["deploymentId"],
  },
})(async (deploymentId: string) => {
  return deployments.get(deploymentId);
});

const agent = new Agent({
  apiKey: process.env.LLM_API_KEY ?? "",
  agentblitApiKey: process.env.AGENTBLIT_API_KEY ?? "",
  customTools: [getDeploymentStatus],
  approvalCallback: async (toolName, args) => {
    return requestApproval(toolName, args);
  },
});

for await (const chunk of agent.run("Check deployment dep_123")) {
  process.stdout.write(chunk);
}

Install the TypeScript SDK with npm install agentblit, then create an agent and API key in the Agentblit console. See the documentation for Python examples, connector setup, and deployment options.