@pmcollab/coworkstream-engine-azure-ai-agent

Coding agents: read packages/workstream-engine/AGENTS.md for the engine integration recipe before wiring this adapter. Migrating an existing setup? See packages/workstream-engine/MIGRATION.md.

Azure AI Agent Service AgentRunner adapter — wraps an agent hosted in Azure AI Foundry.

Install

npm install @pmcollab/coworkstream-engine-azure-ai-agent @azure/ai-projects @azure/identity

Use

import { AIProjectsClient } from '@azure/ai-projects'
import { DefaultAzureCredential } from '@azure/identity'
import { createEngine } from '@pmcollab/coworkstream-engine'
import { createAzureAIAgentAgent } from '@pmcollab/coworkstream-engine-azure-ai-agent'

const projects = AIProjectsClient.fromConnectionString(
  process.env.AI_PROJECT_CONNECTION_STRING,
  new DefaultAzureCredential()
)

const agent = createAzureAIAgentAgent({
  id: 'risk-azure',
  projectsClient: projects,
  agentId: 'asst_xxxx',
})

const engine = createEngine({ agents: [agent], /* ... */ })

Flow per item

1. Create a thread.
2. Post the formatted item as a user message.
3. Create a run against the configured agent.
4. Poll until terminal status (completed, failed, cancelled, expired).
5. Read the assistant's reply and extract { position, reasoning, confidence }.

Recommended agent setup

Configure your Azure AI agent's instructions to end with:

Respond with a JSON object: { "position": "...", "reasoning": "...", "confidence": 0..1 }

The default extractor parses JSON out of the assistant's text. Or set the agent to use structured output (response_format) and the JSON path becomes guaranteed.

Options

createAzureAIAgentAgent({
  id, projectsClient, agentId,    // required
  name?, avatar?, capabilities?,
  pollIntervalMs?: number,        // default 1000
  timeoutMs?: number,             // default 120_000
  defaultConfidence?: number,
  formatItem?: (item, ctx) => string,
  extractPosition?: (message, item, ctx) => { position, reasoning, confidence },
})

License

Commercial. See LICENSE in the repository root.