2
0

docs(ai-service): update readme with plugin protocol integration

Update AI service README to reflect the dual-service architecture (AI operations + plugin protocol) and complete integration guide.

Architecture Updates:
- Document both GitCaddyAIService and PluginService running on port 5000
- Add architecture diagram showing server's AI client and plugin manager connections
- Clarify h2c (cleartext HTTP/2) transport for gRPC + REST on same port

API Reference:
- Add plugin protocol RPC methods (Initialize, HealthCheck, OnEvent, Shutdown)
- Explain plugin lifecycle and health monitoring (30s intervals)
- Document manifest-based capability declaration

Integration Guide:
- Split configuration into [ai] (operations) and [plugins.gitcaddy-ai] (lifecycle)
- Explain how both sections work together for complete integration
- Update client examples to use port 5000 (was 5051)
- Add transport details and event subscription explanation

This provides a complete picture of how the AI service integrates with the server as both an AI operations provider and a managed plugin.
This commit is contained in:
2026-02-13 01:54:41 -05:00
parent ac8aa4c868
commit ea06ca266f

View File

@@ -13,19 +13,30 @@ AI-powered code intelligence service for GitCaddy. Provides code review, documen
## Architecture ## Architecture
The AI service runs as a sidecar alongside the GitCaddy server. It exposes two gRPC services on the same port (HTTP/2 + HTTP/1.1):
- **GitCaddyAIService** - AI operations (review, triage, docs, chat) called by the server's AI client
- **PluginService** - Plugin lifecycle protocol (initialize, health, events) managed by the server's external plugin manager
``` ```
┌─────────────────────────────────────────────────────────────────┐ ┌──────────────────────────────────────────────────────────────────────
│ GitCaddy Server (Go) │ │ GitCaddy Server (Go)
── AI Client (gRPC) ── AI Client (gRPC) ──────────────── GitCaddyAIService
│ ReviewPullRequest, TriageIssue,
ExplainCode, GenerateDocumentation │
┌─────────────────────────────────────────────────────────┐ │ │
│ GitCaddy AI Service (.NET 9) │ └── External Plugin Manager (gRPC) ─── PluginService
├── gRPC API Initialize, HealthCheck (30s), │
├── MarketAlly.AIPlugin (Multi-provider AI) │ OnEvent, Shutdown ▼
│ └── License Validation ┌──────────────────────────┐
└─────────────────────────────────────────────────────────┘ │ GitCaddy AI (.NET 9) │
└─────────────────────────────────────────────────────────────────┘ │ │ Port 5000 (h2c + HTTP) │ │
│ │ ├── AI Providers │ │
│ │ │ (Claude/OpenAI/ │ │
│ │ │ Gemini) │ │
│ │ └── License Validation │ │
│ └──────────────────────────┘ │
└──────────────────────────────────────────────────────────────────────┘
``` ```
## Quick Start ## Quick Start
@@ -94,7 +105,9 @@ Environment variable format: `AIService__DefaultProvider=Claude`
## API Reference ## API Reference
The service exposes a gRPC API defined in `protos/gitcaddy_ai.proto`. The service exposes two gRPC services on port 5000 (cleartext HTTP/2):
### AI Operations (`protos/gitcaddy_ai.proto`)
### Code Review ### Code Review
@@ -134,6 +147,21 @@ rpc ExecuteTask(ExecuteTaskRequest) returns (ExecuteTaskResponse);
rpc Chat(stream ChatRequest) returns (stream ChatResponse); rpc Chat(stream ChatRequest) returns (stream ChatResponse);
``` ```
### Plugin Protocol (`protos/plugin.proto`)
The plugin protocol allows the GitCaddy server to manage this service as an external plugin:
```protobuf
rpc Initialize(InitializeRequest) returns (InitializeResponse);
rpc Shutdown(ShutdownRequest) returns (ShutdownResponse);
rpc HealthCheck(PluginHealthCheckRequest) returns (PluginHealthCheckResponse);
rpc GetManifest(GetManifestRequest) returns (PluginManifest);
rpc OnEvent(PluginEvent) returns (EventResponse);
rpc HandleHTTP(HTTPRequest) returns (HTTPResponse);
```
The server calls `Initialize` on startup, `HealthCheck` every 30 seconds, `OnEvent` for subscribed events (e.g., `license:updated`), and `Shutdown` on server stop. The service declares its capabilities (routes, permissions, license tier) via the `PluginManifest` returned during initialization.
## Client Libraries ## Client Libraries
### .NET Client ### .NET Client
@@ -141,7 +169,7 @@ rpc Chat(stream ChatRequest) returns (stream ChatResponse);
```csharp ```csharp
using GitCaddy.AI.Client; using GitCaddy.AI.Client;
var client = new GitCaddyAIClient("http://localhost:5051"); var client = new GitCaddyAIClient("http://localhost:5000");
var response = await client.ReviewPullRequestAsync(new ReviewPullRequestRequest var response = await client.ReviewPullRequestAsync(new ReviewPullRequestRequest
{ {
@@ -159,7 +187,7 @@ Console.WriteLine(response.Summary);
```go ```go
import ai "git.marketally.com/gitcaddy/gitcaddy-ai/go/gitcaddy-ai-client" import ai "git.marketally.com/gitcaddy/gitcaddy-ai/go/gitcaddy-ai-client"
client, err := ai.NewClient("localhost:5051") client, err := ai.NewClient("localhost:5000")
if err != nil { if err != nil {
log.Fatal(err) log.Fatal(err)
} }
@@ -213,14 +241,43 @@ go generate
## Integration with GitCaddy Server ## Integration with GitCaddy Server
Add the AI client to your GitCaddy server configuration: Add both sections to the server's `app.ini`:
**1. AI client configuration** (how the server calls AI operations):
```ini ```ini
[ai] [ai]
ENABLED = true ENABLED = true
SERVICE_URL = http://localhost:5051 SERVICE_URL = localhost:5000
DEFAULT_PROVIDER = claude
DEFAULT_MODEL = claude-sonnet-4-20250514
CLAUDE_API_KEY = sk-ant-...
``` ```
See the [server README](https://git.marketally.com/gitcaddy/gitcaddy-server#ai-features-configuration) for the full `[ai]` reference.
**2. Plugin registration** (how the server manages the sidecar's lifecycle):
```ini
[plugins]
ENABLED = true
HEALTH_CHECK_INTERVAL = 30s
[plugins.gitcaddy-ai]
ENABLED = true
ADDRESS = localhost:5000
HEALTH_TIMEOUT = 5s
SUBSCRIBED_EVENTS = license:updated
```
With both sections configured, the server will:
- Call AI operations (review, triage, etc.) via `[ai] SERVICE_URL`
- Manage the sidecar's lifecycle via the plugin protocol on `[plugins.gitcaddy-ai] ADDRESS`
- Health-check the sidecar every 30 seconds and log status changes
- Dispatch subscribed events (e.g., license updates) to the sidecar in real-time
**Transport**: All communication uses cleartext HTTP/2 (h2c). The sidecar's Kestrel server is configured for `Http1AndHttp2` on port 5000, supporting both gRPC (HTTP/2) and REST (HTTP/1.1) on the same port.
## Support ## Support
- Documentation: https://docs.gitcaddy.com/ai - Documentation: https://docs.gitcaddy.com/ai