GoClaw/server/gateway-proxy.ts

/**
 * GoClaw Gateway Proxy
 *
 * Forwards orchestrator/agent/tool requests from the Node.js tRPC server
 * to the Go Gateway running on :18789.
 *
 * The Go Gateway handles:
 *   - LLM orchestration (tool-use loop)
 *   - Tool execution (shell, file, docker, http)
 *   - Agent listing from DB
 *   - Model listing from LLM provider
 *
 * When GATEWAY_URL is not set or the gateway is unreachable, all functions
 * return a structured error so callers can fall back gracefully.
 */

const GATEWAY_BASE_URL = process.env.GATEWAY_URL ?? "http://localhost:18789";
const GATEWAY_TIMEOUT_MS = 180_000; // 3 min — LLM can be slow
const QUICK_TIMEOUT_MS = 5_000;

// ─── Types ────────────────────────────────────────────────────────────────────

export interface GatewayMessage {
  role: "user" | "assistant" | "system";
  content: string;
}

export interface GatewayToolCallStep {
  tool: string;
  args: Record<string, unknown>;
  result: unknown; // required — matches ToolCallStep interface in Chat.tsx / Skills.tsx
  error?: string;
  success: boolean;
  durationMs: number;
}

export interface GatewayChatResult {
  success: boolean;
  response: string;
  toolCalls: GatewayToolCallStep[];
  model?: string;
  modelWarning?: string;
  usage?: {
    prompt_tokens: number;
    completion_tokens: number;
    total_tokens: number;
  };
  error?: string;
}

export interface GatewayOrchestratorConfig {
  id: number | null;
  name: string;
  model: string;
  temperature: number;
  maxTokens: number;
  allowedTools: string[];
  systemPromptPreview: string;
}

export interface GatewayToolDef {
  name: string;
  description: string;
  parameters: Record<string, unknown>;
}

export interface GatewayToolResult {
  success: boolean;
  output?: unknown;
  error?: string;
  durationMs: number;
}

export interface GatewayHealthResult {
  connected: boolean;
  latencyMs: number;
  llm?: { connected: boolean; latencyMs: number };
  error?: string;
}

// ─── Health ───────────────────────────────────────────────────────────────────

/**
 * Check if the Go Gateway is running and healthy.
 * Also returns LLM provider health (Ollama / cloud API).
 */
export async function checkGatewayHealth(): Promise<GatewayHealthResult> {
  const start = Date.now();
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/health`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    const latencyMs = Date.now() - start;
    if (res.ok) {
      const data = await res.json();
      return {
        connected: true,
        latencyMs,
        // Go Gateway may return "ollama" or "llm" key depending on version
        llm: data.llm ?? data.ollama,
      };
    }
    return { connected: false, latencyMs, error: `HTTP ${res.status}` };
  } catch (err: unknown) {
    const msg = err instanceof Error ? err.message : String(err);
    return { connected: false, latencyMs: Date.now() - start, error: msg };
  }
}

/**
 * Returns true if the Go Gateway is reachable (fast check, no LLM ping).
 */
export async function isGatewayAvailable(): Promise<boolean> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/health`, {
      signal: AbortSignal.timeout(2_000),
    });
    return res.ok;
  } catch {
    return false;
  }
}

// ─── Model Info ───────────────────────────────────────────────────────────────

export interface OllamaModelInfo {
  contextLength: number;
  parameterSize?: string;
  family?: string;
  quantization?: string;
  capabilities?: string[];
}

/**
 * Fetch model details from Ollama /api/show (context_length, parameters, etc.)
 * Uses the base URL and API key from environment.
 */
export async function getOllamaModelInfo(modelId: string): Promise<OllamaModelInfo | null> {
  const baseUrl = (process.env.OLLAMA_BASE_URL ?? "https://ollama.com/v1").replace(/\/v1\/?$/, "");
  const apiKey = process.env.OLLAMA_API_KEY ?? "";
  try {
    const headers: Record<string, string> = { "Content-Type": "application/json" };
    if (apiKey) headers["Authorization"] = `Bearer ${apiKey}`;
    const res = await fetch(`${baseUrl}/api/show`, {
      method: "POST",
      headers,
      body: JSON.stringify({ model: modelId }),
      signal: AbortSignal.timeout(10_000),
    });
    if (!res.ok) return null;
    const data = await res.json();
    // context_length is under model_info with key "{arch}.context_length"
    let contextLength = 0;
    if (data.model_info) {
      for (const [k, v] of Object.entries(data.model_info)) {
        if (k.endsWith(".context_length") && typeof v === "number") {
          contextLength = v;
          break;
        }
      }
    }
    return {
      contextLength,
      parameterSize: data.details?.parameter_size,
      family: data.details?.family,
      quantization: data.details?.quantization_level,
      capabilities: data.capabilities,
    };
  } catch {
    return null;
  }
}

// ─── Orchestrator ─────────────────────────────────────────────────────────────

/**
 * Send a chat message to the Go Orchestrator.
 * Includes full tool-use loop — the Go side handles all LLM ↔ tool iterations.
 */
export async function gatewayChat(
  messages: GatewayMessage[],
  model?: string,
  maxIter?: number
): Promise<GatewayChatResult> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/orchestrator/chat`, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ messages, model, maxIter }),
      signal: AbortSignal.timeout(GATEWAY_TIMEOUT_MS),
    });
    if (!res.ok) {
      const text = await res.text();
      return {
        success: false,
        response: "",
        toolCalls: [],
        error: `Gateway error (${res.status}): ${text}`,
      };
    }
    return res.json();
  } catch (err: unknown) {
    const msg = err instanceof Error ? err.message : String(err);
    return {
      success: false,
      response: "",
      toolCalls: [],
      error: `Gateway unreachable: ${msg}. Is the Go Gateway running on ${GATEWAY_BASE_URL}?`,
    };
  }
}

/**
 * Get orchestrator config from Go Gateway (reads from DB).
 */
export async function getGatewayOrchestratorConfig(): Promise<GatewayOrchestratorConfig | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/orchestrator/config`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

// ─── Models ───────────────────────────────────────────────────────────────────

/**
 * Get list of models from Go Gateway (proxied from LLM provider).
 */
export async function getGatewayModels(): Promise<{ data: { id: string }[] } | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/models`, {
      signal: AbortSignal.timeout(10_000),
    });
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

// ─── Agents ───────────────────────────────────────────────────────────────────

/**
 * Get list of agents from Go Gateway (reads from DB).
 */
export async function getGatewayAgents(): Promise<{ agents: unknown[]; count: number } | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/agents`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

/**
 * Get a single agent by ID from Go Gateway.
 */
export async function getGatewayAgent(id: number): Promise<unknown | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/agents/${id}`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

// ─── Tools ────────────────────────────────────────────────────────────────────

/**
 * Get list of available tools from Go Gateway.
 */
export async function getGatewayTools(): Promise<GatewayToolDef[] | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/tools`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    if (!res.ok) return null;
    const data = await res.json();
    // Go returns OpenAI format: { tools: [{type: "function", function: {name, description, parameters}}, ...], count: N }
    if (!Array.isArray(data.tools)) return null;
    return data.tools.map((t: { type?: string; function?: { name: string; description: string; parameters: Record<string, unknown> }; name?: string; description?: string; parameters?: Record<string, unknown> }) => {
      // Handle OpenAI format: {type: "function", function: {name, ...}}
      if (t.function) {
        return {
          name: t.function.name,
          description: t.function.description,
          parameters: t.function.parameters,
        } as GatewayToolDef;
      }
      // Handle flat format: {name, description, parameters}
      return t as GatewayToolDef;
    });
  } catch {
    return null;
  }
}

/**
 * Execute a single tool via Go Gateway.
 */
export async function executeGatewayTool(
  toolName: string,
  args: Record<string, unknown>
): Promise<GatewayToolResult> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/tools/execute`, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ tool: toolName, args }),
      signal: AbortSignal.timeout(30_000),
    });
    if (!res.ok) {
      const text = await res.text();
      return {
        success: false,
        error: `Gateway error (${res.status}): ${text}`,
        durationMs: 0,
      };
    }
    return res.json();
  } catch (err: unknown) {
    const msg = err instanceof Error ? err.message : String(err);
    return {
      success: false,
      error: `Gateway unreachable: ${msg}`,
      durationMs: 0,
    };
  }
}

// ─── Nodes / Docker Swarm ─────────────────────────────────────────────────────

export interface GatewayNodeInfo {
  id: string;
  hostname: string;
  role: string;
  status: string;
  availability: string;
  ip: string;
  os: string;
  arch: string;
  cpuCores: number;
  memTotalMB: number;
  dockerVersion: string;
  isLeader: boolean;
  managerAddr?: string;
  labels: Record<string, string>;
  updatedAt: string;
}

export interface GatewayContainerInfo {
  id: string;
  name: string;
  image: string;
  state: string;
  status: string;
}

export interface GatewayNodesResult {
  nodes: GatewayNodeInfo[];
  count: number;
  swarmActive: boolean;
  managers?: number;
  totalNodes?: number;
  containers?: GatewayContainerInfo[];
  fetchedAt: string;
}

export interface GatewayContainerStat {
  id: string;
  name: string;
  cpuPct: number;
  memUseMB: number;
  memLimMB: number;
  memPct: number;
}

export interface GatewayNodeStatsResult {
  stats: GatewayContainerStat[];
  count: number;
  fetchedAt: string;
}

/**
 * Get Docker Swarm nodes (or standalone Docker host) from Go Gateway.
 */
export async function getGatewayNodes(): Promise<GatewayNodesResult | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/nodes`, {
      signal: AbortSignal.timeout(10_000),
    });
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

/**
 * Get live container CPU/RAM stats from Go Gateway.
 */
export async function getGatewayNodeStats(): Promise<GatewayNodeStatsResult | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/nodes/stats`, {
      signal: AbortSignal.timeout(15_000),
    });
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

// ─── Persistent Chat Sessions ─────────────────────────────────────────────────

export interface GatewayChatEvent {
  id: number;
  sessionId: string;
  seq: number;
  eventType: "thinking" | "tool_call" | "delta" | "done" | "error";
  content: string;
  toolName: string;
  toolArgs: string;   // JSON string
  toolResult: string;
  toolSuccess: boolean;
  durationMs: number;
  model: string;
  usageJson: string;  // JSON string
  errorMsg: string;
  createdAt: string;
}

export interface GatewayChatSession {
  id: number;
  sessionId: string;
  agentId: number;
  status: "running" | "done" | "error";
  userMessage: string;
  finalResponse: string;
  model: string;
  totalTokens: number;
  processingTimeMs: number;
  errorMessage: string;
  createdAt: string;
  updatedAt: string;
}

/**
 * Start a persistent background chat session.
 * Returns the sessionId immediately; processing continues on the server.
 */
export async function startChatSession(
  messages: GatewayMessage[],
  sessionId: string,
  model?: string,
  maxIter = 10
): Promise<{ sessionId: string; status: string } | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/chat/session`, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ messages, sessionId, model, maxIter }),
      signal: AbortSignal.timeout(10_000),
    });
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

/**
 * Get session metadata (status, finalResponse, tokens…).
 */
export async function getChatSession(sessionId: string): Promise<GatewayChatSession | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/chat/session/${sessionId}`, {
      signal: AbortSignal.timeout(5_000),
    });
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

/**
 * Fetch events for a session with seq > afterSeq.
 * Returns { sessionId, status, events[] }.
 */
export async function getChatEvents(
  sessionId: string,
  afterSeq = 0
): Promise<{ sessionId: string; status: string; events: GatewayChatEvent[] } | null> {
  try {
    const res = await fetch(
      `${GATEWAY_BASE_URL}/api/chat/session/${sessionId}/events?after=${afterSeq}`,
      { signal: AbortSignal.timeout(5_000) }
    );
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

/**
 * List recent sessions (default last 50).
 */
export async function listChatSessions(
  limit = 50
): Promise<{ sessions: GatewayChatSession[] } | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/chat/sessions?limit=${limit}`, {
      signal: AbortSignal.timeout(5_000),
    });
    if (!res.ok) return null;
    return res.json();
  } catch {
    return null;
  }
}

// ─── Real Docker Swarm API ────────────────────────────────────────────────────

export interface SwarmNodeInfo {
  id: string;
  hostname: string;
  role: "manager" | "worker";
  state: string;
  availability: string;
  ip: string;
  os: string;
  arch: string;
  cpuCores: number;
  memTotalMB: number;
  dockerVersion: string;
  isLeader: boolean;
  managerAddr?: string;
  labels: Record<string, string>;
  updatedAt: string;
}

export interface SwarmServiceInfo {
  id: string;
  name: string;
  image: string;
  mode: "replicated" | "global";
  desiredReplicas: number;
  runningTasks: number;
  desiredTasks: number;
  labels: Record<string, string>;
  updatedAt: string;
  ports: string[];
  isGoClaw: boolean;
}

export interface SwarmTaskInfo {
  id: string;
  serviceId: string;
  nodeId: string;
  slot: number;
  state: string;
  message: string;
  containerId: string;
  updatedAt: string;
}

export interface SwarmInfoResult {
  nodeId: string;
  localNodeState: string;
  isManager: boolean;
  managers: number;
  nodes: number;
  managerAddr: string;
  joinTokens?: { worker: string; manager: string };
}

export interface JoinTokenResult {
  role: string;
  token: string;
  managerAddr: string;
  joinCommand: string;
}

/** Get overall swarm state + join tokens */
export async function getSwarmInfo(): Promise<SwarmInfoResult | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/info`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    if (!res.ok) return null;
    return res.json();
  } catch { return null; }
}

/** List all swarm nodes with live status */
export async function listSwarmNodes(): Promise<{ nodes: SwarmNodeInfo[]; count: number } | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/nodes`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    if (!res.ok) return null;
    return res.json();
  } catch { return null; }
}

/** List all swarm services */
export async function listSwarmServices(): Promise<{ services: SwarmServiceInfo[]; count: number } | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/services`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    if (!res.ok) return null;
    return res.json();
  } catch { return null; }
}

/** Get tasks for a specific service */
export async function getServiceTasks(serviceId: string): Promise<{ tasks: SwarmTaskInfo[] } | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/services/${serviceId}/tasks`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    if (!res.ok) return null;
    return res.json();
  } catch { return null; }
}

/** Scale a service to N replicas */
export async function scaleSwarmService(serviceId: string, replicas: number): Promise<boolean> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/services/${serviceId}/scale`, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ replicas }),
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    return res.ok;
  } catch { return false; }
}

/** Get join token and command */
export async function getSwarmJoinToken(role: "worker" | "manager"): Promise<JoinTokenResult | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/join-token?role=${role}`, {
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    if (!res.ok) return null;
    return res.json();
  } catch { return null; }
}

/** Execute a shell command on the host system */
export async function execSwarmShell(command: string): Promise<{ output: string; success: boolean; error?: string } | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/shell`, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ command }),
      signal: AbortSignal.timeout(35_000),
    });
    if (!res.ok) return null;
    return res.json();
  } catch { return null; }
}

/** Add a label to a swarm node */
export async function addSwarmNodeLabel(nodeId: string, key: string, value: string): Promise<boolean> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/nodes/${nodeId}/label`, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ key, value }),
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    return res.ok;
  } catch { return false; }
}

/** Set node availability (active|pause|drain) */
export async function setNodeAvailability(nodeId: string, availability: "active" | "pause" | "drain"): Promise<boolean> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/nodes/${nodeId}/availability`, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ availability }),
      signal: AbortSignal.timeout(QUICK_TIMEOUT_MS),
    });
    return res.ok;
  } catch { return false; }
}

/** Deploy a new agent as a Swarm service */
export async function createAgentService(opts: {
  name: string; image: string; replicas: number; env?: string[]; port?: number;
}): Promise<{ ok: boolean; serviceId?: string; name?: string } | null> {
  try {
    const res = await fetch(`${GATEWAY_BASE_URL}/api/swarm/services/create`, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify(opts),
      signal: AbortSignal.timeout(15_000),
    });
    if (!res.ok) return null;
    return res.json();
  } catch { return null; }
}