Agent SDK (@solenai/sdk)

Three lines of code. Wrap any agent — LangGraph, CrewAI, AutoGen, or custom — and Solen checkpoints state, monitors every step, and heals failures in production.

Install

Copy-paste this into your agent entry point:

npm install @solenai/sdk

import { withSolen } from '@solenai/sdk';

await withSolen(
  {
    apiKey: process.env.SOLEN_API_KEY!,
    agentId: 'research-agent',
    framework: 'langgraph',
  },
  async (ctx) => {
    // Your agent logic here
    return result;
  }
);

Get your apiKey from Settings → API Keys.

What happens after install

Once withSolen() runs:

1. Solen creates a run via POST /api/agent-runtime/runs with your agentId and framework
2. Every ctx.step() reports tool calls, LLM invocations, duration, and status to the Runtime Kernel
3. Checkpoints are saved at configurable intervals (default every 10 steps) and on explicit ctx.checkpoint() calls
4. On step failure or stall, Solen opens a heal incident and either pattern-matches a known fix or runs AI diagnosis
5. On completion, the run is marked COMPLETED or FAILED with failure detail

The SDK never throws on network failures — telemetry is fire-and-forget so it cannot crash your agent.

Reporting steps

import { withSolen } from '@solenai/sdk';

await withSolen(config, async (ctx) => {
  const data = await ctx.step('tool_call', () => searchWeb(query), {
    toolName: 'search_web',
  });
  const analysis = await ctx.step('llm_call', () => analyzeWithLlm(data));
  return analysis;
});

Checkpoints

import { withSolen } from '@solenai/sdk';

await withSolen(config, async (ctx) => {
  const state = { messages, currentStep: 10 };
  await ctx.checkpoint(state);
  return state;
});

Configuration options

Requirements

• Node.js 18+ (uses built-in fetch)
• Zero runtime dependencies
• Works with LangGraph, CrewAI, AutoGen, and custom agent loops

← All integrations