agent-registry-mongo

@computeragent/agent-registry-mongo

MongoDB-backed agent registry + per-run audit log + AgentTelemetry implementation for ComputerAgent.

Use this when ComputerAgent runs as a library inside an existing worker (e.g. a Temporal worker pod, a CLI batch job, a serverless function) and you want the AgentOS dashboard to see every run with no extra orchestration. Pass the telemetry instance to the SDK, and the SDK fires lifecycle hooks that upsert to agent_registry and append to agent_logs.

Install

npm install @computeragent/agent-registry-mongo

Use

import { ComputerAgent, LocalSubstrate } from "computeragent";
import { MongoTelemetry } from "@computeragent/agent-registry-mongo";

const telemetry = new MongoTelemetry({
  url: process.env.MONGO_URL!,
  database: "agentos",
  agent: {
    name: "devsupport-agent",
    label: "DevSupport",
    source: "github.com/nord/devsupport-agent",
    harness: "claude-agent-sdk",
    model: "bedrock/anthropic.claude-sonnet-4-20250514-v1:0",
    registeredBy: process.env.HOSTNAME, // pod name, "ci", etc.
  },
});

await using agent = new ComputerAgent({
  source: {type: "git", url: "github.com/nord/devsupport-agent"},
  harness: "claude-agent-sdk",
  runtime: new LocalSubstrate(),
  envs: { CLAUDE_CODE_USE_BEDROCK: "1", AWS_REGION: "us-west-2" },
  telemetry,
});

const result = await agent.chat("hello");

That's it. On construct, an upsert lands in agent_registry. On every chat(), a document lands in agent_logs with usage, durationMs, ok, error, and a truncated reply.

Collections

agent_registry

{
  _id: "devsupport-agent",          // agent name (unique)
  label: "DevSupport",
  harness: "claude-agent-sdk",
  source: "github.com/nord/...",    // IdentitySource JSON
  model: "bedrock/anthropic.claude-sonnet-4-...",
  registeredBy: "worker-abc-123",
  registeredAt: ISODate("..."),
  updatedAt:    ISODate("..."),
  lastSeen:     ISODate("..."),
}

agent_logs

{
  _id: "log_<uuid>",
  ts: ISODate("..."),
  source: "library",                // | "slack" | "web" | "schedule" | ...
  agentName: "devsupport-agent",
  sessionId: "session_...",
  query: "user message text",
  reply: "assistant reply text",
  ok: true,
  durationMs: 12345,
  inputTokens: 1200,
  outputTokens: 350,
  costUsd: 0.012,
}

Lower-level API

If you want CRUD beyond the telemetry hook (e.g. unregistering an agent from a teardown script):

import { AgentRegistry, AgentLogStore } from "@computeragent/agent-registry-mongo";

const reg = new AgentRegistry({url, database: "agentos"});
await reg.register({name: "agent-1", harness: "claude-agent-sdk", source: "..."});
await reg.list();
await reg.unregister("agent-1");
await reg.close();

const logs = new AgentLogStore({url, database: "agentos"});
const recent = await logs.list({agentName: "agent-1", limit: 20});

Connection sharing

If your app already manages a MongoClient, pass it in to avoid a second pool:

const client = new MongoClient(process.env.MONGO_URL!);
await client.connect();
const telemetry = new MongoTelemetry({
  url: process.env.MONGO_URL!,
  database: "agentos",
  agent: {name: "devsupport-agent", harness: "claude-agent-sdk", source: "..."},
  client,                            // shared
});
// telemetry.onClose() won't close `client` — you own its lifecycle.

Error policy

Telemetry calls are fire-and-forget on the SDK side — exceptions thrown here never reach the agent's chat result. For diagnostics, pass an onError callback:

new MongoTelemetry({
  ...,
  onError: (err, op) => myLogger.warn({err, op}, "telemetry write failed"),
});