kairos-sdk

The flight recorder for autonomous operations.

Instrument any AI agent in 3 lines. Every prompt, decision, tool call, and failure — captured, replayable, and governable.

Install

npm install kairos-sdk

Quickstart

import { createKairos } from 'kairos-sdk'

const kairos = createKairos()

const exec = kairos.execution({ workflowName: 'research-agent' })

exec.setPrompt(userPrompt)
exec.setModel('claude-sonnet-4-6')
exec.toolCall({ name: 'web_search', input: { query }, output: results, latencyMs: 1200 })
exec.decision('Selected most credible source', 0.91)

await exec.complete(output)

That's it. Open kairos-web-lyart.vercel.app/app to see the execution.

API

createKairos(config?)

const kairos = createKairos({
  baseUrl: 'https://kairos-production-64c5.up.railway.app', // default
  debug: false,
})

kairos.execution(init?)

Returns a KairosExecution builder. All methods are chainable.

| Method | Description | |--------|-------------| | .setPrompt(prompt) | Record the prompt sent to the model | | .setModel(model) | Set the model name | | .setTokens(prompt, completion) | Set token counts | | .setCost(usd) | Set cost in USD | | .toolCall({ name, input, output, latencyMs }) | Record a tool call (auto-emits tool_called + tool_completed) | | .decision(reasoning, confidence?) | Record a model decision | | .policyCheck(policy, result) | Record a policy/governance check | | .memoryWrite(key, value?) | Record a memory write | | .memoryRead(key, value?) | Record a memory read | | .retry(attempt, reason?) | Record a retry | | .event(type, payload?) | Emit any custom event | | .complete(output?) | Finish execution as completed | | .fail(error) | Finish execution as failed |

Events

All standard event types are supported with full autocomplete:

workflow_started · prompt_sent · tool_called · tool_completed · tool_failed · decision_made · policy_checked · memory_written · memory_read · approval_requested · retry_triggered · execution_completed · execution_failed

License

MIT