@opensearch-project/agent-health 0.3.0 → 0.5.2

package/docs/rfcs/003-remote-aggregation.md

@@ -0,0 +1,146 @@
+<!--
+  * Copyright OpenSearch Contributors
+  * SPDX-License-Identifier: Apache-2.0
+-->
+
+# RFC 003: Remote Server Aggregation
+
+| Field        | Value                                      |
+|--------------|--------------------------------------------|
+| **Status**   | Implemented                                |
+| **Author**   | Anirudha Jadhav                            |
+| **Created**  | 2026-04-07                                 |
+| **Depends**  | RFC 001 (Coding Agent Analytics)           |
+| **Issue**    | opensearch-project/agent-health#122        |
+
+## Summary
+
+Extend Agent Health's Coding Agent Analytics (RFC 001) with remote server aggregation. Developers running coding agents on multiple machines (laptops, EC2 build servers, remote VMs) can aggregate session data into a single local dashboard. Each remote machine runs Agent Health in headless mode with API key authentication, and the local instance transparently merges remote data with local sessions.
+
+## Motivation
+
+### Problem
+
+1. **Multi-machine development** — Developers increasingly use remote build servers, cloud dev environments, and multiple laptops. Session data is siloed per machine.
+2. **Team visibility** — Engineering leads want to see aggregated usage across their team's machines without requiring each developer to manually export data.
+3. **No standard aggregation** — Existing tools (cc-lens, vibedev) only read local filesystem data. There is no way to pull coding agent analytics from remote machines.
+
+### Why Extend RFC 001
+
+RFC 001 established the reader interface, session cache, and analytics API. Remote aggregation reuses the same `CodingAgentReader` interface and API surface — the `RemoteAggregator` extends `CodingAgentRegistry` and merges remote sessions transparently. No changes to the frontend are needed beyond server name badges on sessions.
+
+## Design
+
+### Architecture
+
+```
+┌─────────────┐     ┌─────────────┐     ┌─────────────┐
+│  EC2 Build  │     │  Laptop 2   │     │  Cloud Dev  │
+│  Server     │     │  (headless) │     │  (headless) │
+│  :4001      │     │  :4001      │     │  :4001      │
+└──────┬──────┘     └──────┬──────┘     └──────┬──────┘
+       │                   │                   │
+       └──────────┬────────┴───────────────────┘
+                  │  HTTP + API Key
+           ┌──────┴──────┐
+           │  Local Dev  │
+           │  Dashboard  │
+           │  :4001      │
+           └─────────────┘
+```
+
+### Remote Server Mode
+
+Remote machines run Agent Health in headless mode:
+
+```bash
+npx @opensearch-project/agent-health serve --headless --api-key sk-my-secret
+```
+
+- No UI served, only API endpoints
+- API key required for all `/api/coding-agents/*` routes
+- Responds to the same analytics endpoints as local
+
+### RemoteAggregator
+
+The `RemoteAggregator` class extends `CodingAgentRegistry`:
+
+- On each API call, fetches from all configured remote servers in parallel
+- Tags each remote session with `server_name` for attribution
+- Uses `Promise.allSettled` for fault isolation — one slow/failing remote doesn't block local data
+- 30-second cache TTL for remote data, 10-second request timeout
+- Graceful degradation: if a remote is unreachable, local data still serves
+
+### API Key Authentication
+
+- Middleware checks `Authorization: Bearer <key>` header
+- Only applied to `/api/coding-agents/*` routes when `--api-key` is set
+- No-op when no API key configured (local-only mode)
+- Keys stored in `agent-health.config.json` per remote server
+
+### Configuration
+
+Remote servers are configured via `agent-health.config.json`:
+
+```json
+{
+  "remoteServers": [
+    { "name": "ec2-build-1", "url": "http://10.0.1.50:4001", "apiKey": "sk-abc123" }
+  ]
+}
+```
+
+Or via CLI:
+
+```bash
+agent-health remote add --name ec2-build-1 --url http://10.0.1.50:4001 --api-key sk-abc
+agent-health remote remove ec2-build-1
+agent-health remote list
+agent-health remote test
+```
+
+### Settings UI
+
+A Settings page allows managing remote servers from the browser:
+- Add/remove remote servers
+- Test connectivity
+- View connection status
+- API keys masked in display
+
+## API Endpoints
+
+No new public endpoints. The existing `/api/coding-agents/*` endpoints transparently include remote data when remote servers are configured. Internal endpoints:
+
+| Endpoint | Method | Purpose |
+|----------|--------|---------|
+| `/api/remote-servers` | GET | List configured servers (keys masked) |
+| `/api/remote-servers` | POST | Add server |
+| `/api/remote-servers/:name` | DELETE | Remove server |
+| `/api/remote-servers/:name/test` | POST | Test connectivity |
+
+## Security Considerations
+
+- API keys are stored in a local config file, not in environment variables
+- Keys are masked in API responses (only last 4 chars shown)
+- Remote connections should use HTTPS in production
+- Rate limiting is recommended for exposed endpoints
+- No raw conversation data is served over the remote protocol — only aggregated session metadata
+
+## Files
+
+| File | Purpose |
+|------|---------|
+| `server/services/codingAgents/remoteAggregator.ts` | RemoteAggregator class |
+| `server/services/codingAgents/remoteConfig.ts` | Config file management |
+| `server/middleware/apiKeyAuth.ts` | API key auth middleware |
+| `server/routes/config.ts` | Remote server management routes |
+| `cli/commands/remote.ts` | CLI remote commands |
+| `components/SettingsPage.tsx` | Settings UI |
+| `tests/unit/server/middleware/apiKeyAuth.test.ts` | Auth middleware tests |
+| `tests/unit/server/services/codingAgents/remoteAggregator.test.ts` | Aggregator tests |
+| `tests/unit/server/services/codingAgents/remoteConfig.test.ts` | Config tests |
+
+## Relationship to Other RFCs
+
+- **Depends on RFC 001**: Uses reader interface, session cache, registry pattern
+- **Complements RFC 002**: Remote aggregation provides cross-machine data that the enterprise leaderboard can ingest. However, as noted in the RFC 002 review, OpenSearch sync may eventually replace machine-to-machine HTTP for enterprise use cases.