REST Endpoints

Debug API

Method	Path	Description
GET	/debug/api/	List all debug entries (summaries)
GET	/debug/api/summary/{id}	Single entry summary
GET	/debug/api/view/{id}	Full entry data (optionally filtered by collector)
GET	/debug/api/dump/{id}	Dump objects for entry
GET	/debug/api/object/{id}/{objectId}	Specific object from dump
GET	/debug/api/settings	Debug settings (path mapping)

Example: List entries

GET /debug/api/

{
    "id": null,
    "data": [
        {
            "id": "abc123",
            "collectors": ["request", "log", "event"],
            "url": "/api/users",
            "method": "GET",
            "status": 200,
            "time": 1234567890
        }
    ],
    "error": null,
    "success": true
}

Example: View entry

GET /debug/api/view/abc123?collector=log

Returns full collected data for the specified entry, optionally filtered to a single collector.

OTLP Trace Ingestion

Method	Path	Description
POST	/debug/api/otlp/v1/traces	Ingest OpenTelemetry traces in OTLP format

LLM API

All LLM endpoints share the same settings stored in .llm-settings.json. Both the debug panel and the toolbar use these endpoints — configure once, use everywhere.

See the AI Chat guide for user-facing documentation.

LLM Providers

Provider	Auth	How it works
openrouter	API key or OAuth	HTTP calls to OpenRouter API (default)
anthropic	API key or OAuth token	HTTP calls to Anthropic Messages API
openai	API key	HTTP calls to OpenAI Chat Completions API
acp	None (local agent)	Spawns agent (e.g. Claude Code) as subprocess via Agent Client Protocol

Connection

Method	Path	Description
GET	/debug/api/llm/status	Connection status, provider, model, timeout, custom prompt
POST	/debug/api/llm/connect	Connect with API key or ACP agent
POST	/debug/api/llm/disconnect	Clear stored credentials

Status

GET /debug/api/llm/status

{
    "data": {
        "connected": true,
        "provider": "openrouter",
        "model": "anthropic/claude-sonnet-4",
        "timeout": 30,
        "customPrompt": "Reply in English. Be concise and actionable..."
    }
}

Connect with API Key

POST /debug/api/llm/connect
Content-Type: application/json

{"provider": "anthropic", "apiKey": "sk-ant-..."}

Supported providers: openrouter, anthropic, openai.

Connect with ACP

The ACP provider uses the Agent Client Protocol to communicate with local AI agents over stdio. ADP runs a persistent daemon process that manages multiple agent subprocesses — one per browser tab (session).

POST /debug/api/llm/connect
Content-Type: application/json
X-Acp-Session: <uuid>

{
  "provider": "acp",
  "acpCommand": "npx",
  "acpArgs": ["@agentclientprotocol/claude-agent-acp"]
}

• acpCommand — Executable to run (default: npx). Must be on system PATH.
• acpArgs — Arguments passed to the command (default: ["@agentclientprotocol/claude-agent-acp"]).
• acpEnv — Environment variables merged into the agent's environment.

Required header: X-Acp-Session — a UUID identifying the browser tab. Generated via crypto.randomUUID() and stored in sessionStorage. The frontend SDK injects this header automatically on all LLM API requests.

Connect flow:

1. Start daemon (if not running, or restart if protocol version mismatch)
2. Start agent session for the given X-Acp-Session UUID
3. Save settings only after both steps succeed

Response:

{
  "data": {
    "connected": true,
    "provider": "acp",
    "acpCommand": "npx",
    "sessionId": "550e8400-e29b-41d4-a716-446655440000",
    "agentName": "Claude",
    "agentVersion": "1.0.0"
  }
}

ACP adapters bridge the protocol to specific agents. Known adapters:

• @agentclientprotocol/claude-agent-acp — Claude Code (default)
• @anthropic-ai/gemini-agent-acp — Gemini CLI
• @anthropic-ai/codex-agent-acp — Codex CLI

Daemon architecture:

The daemon (acp-daemon-runner.php) is a standalone PHP process that listens on a Unix socket at /tmp/adp-acp-{hash}.sock. It manages agent subprocesses via an internal protocol:

Action	Description
ping	Health check, returns {ok, protocol, sessions}
session-start	Spawn agent subprocess, perform ACP initialize handshake
session-stop	Terminate a specific agent
session-status	Check if a session's agent is alive
prompt	Route session/new + session/prompt to the correct agent
shutdown	Stop all sessions and exit

Limits: max 10 concurrent sessions, 30-minute idle timeout per session.

Protocol lifecycle per chat request: initialize (on session start) → session/new → session/prompt (with streaming session/update notifications) → response.

OAuth (OpenRouter)

Method	Path	Description
POST	/debug/api/llm/oauth/initiate	Start OAuth PKCE flow, returns authUrl and codeVerifier
POST	/debug/api/llm/oauth/exchange	Exchange authorization code + codeVerifier for API key

Initiate

POST /debug/api/llm/oauth/initiate
Content-Type: application/json

{"callbackUrl": "https://localhost:5173/debug/llm/callback"}

{
    "data": {
        "authUrl": "https://openrouter.ai/auth?...",
        "codeVerifier": "..."
    }
}

Open authUrl in a popup. After user approval, OpenRouter redirects to callbackUrl?code=.... Exchange the code:

POST /debug/api/llm/oauth/exchange
Content-Type: application/json

{"code": "...", "codeVerifier": "..."}

Settings

Method	Path	Description
GET	/debug/api/llm/models	List available models for the connected provider
POST	/debug/api/llm/model	Set active model: {"model": "anthropic/claude-sonnet-4"}
POST	/debug/api/llm/timeout	Set timeout in seconds (5–300): {"timeout": 60}
POST	/debug/api/llm/custom-prompt	Set system prompt: {"customPrompt": "..."}

All settings endpoints return the updated LlmStatus object (same shape as GET /status).

Chat & Analysis

Method	Path	Description
POST	/debug/api/llm/chat	Send chat completion request
POST	/debug/api/llm/analyze	Analyze debug entry data with AI

Chat

POST /debug/api/llm/chat
Content-Type: application/json

{
    "messages": [
        {"role": "user", "content": "Why is my /api/users endpoint slow?"}
    ],
    "model": "anthropic/claude-sonnet-4",
    "temperature": 0.7
}

{
    "data": {
        "choices": [{"message": {"role": "assistant", "content": "Based on..."}}],
        "model": "anthropic/claude-sonnet-4",
        "usage": {"prompt_tokens": 50, "completion_tokens": 200}
    }
}

The custom prompt (if set) is automatically prepended as a system message. All responses are normalized to the OpenAI-compatible format regardless of provider.

Analyze

POST /debug/api/llm/analyze
Content-Type: application/json

{
    "context": {"request": {"method": "GET", "path": "/api/users"}, "db": {"queries": {"total": 47}}},
    "prompt": "Why are there so many queries?"
}

Context is truncated to 12,000 characters. Returns {"data": {"analysis": "...", "model": "..."}}.

History

Method	Path	Description
GET	/debug/api/llm/history	Get all history entries (max 100, FIFO)
POST	/debug/api/llm/history	Add entry: {query, response, timestamp, error?}
DELETE	/debug/api/llm/history/{index}	Delete entry by index
DELETE	/debug/api/llm/history	Clear all history

History is shared between the panel and toolbar — entries added from either appear in both.

Ingestion API

Method	Path	Description
POST	/debug/api/ingest/	Ingest single debug entry
POST	/debug/api/ingest/batch	Ingest multiple entries
POST	/debug/api/ingest/log	Shorthand: ingest a single log entry
GET	/debug/api/ingest/openapi.json	OpenAPI 3.1 specification

Service Registry API

Method	Path	Description
POST	/debug/api/services/register	Register an external service
POST	/debug/api/services/heartbeat	Keep service online
GET	/debug/api/services/	List registered services
DELETE	/debug/api/services/{service}	Deregister a service