feat(exec): add LiteLLM HTTP executor for local model dispatch

internal/exec/litellm.go

@@ -0,0 +1,103 @@
+package exec
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"time"
+)
+
+// LiteLLMExecutor calls a LiteLLM-compatible /v1/chat/completions endpoint.
+// Local models are expected to return a JSON object matching the Result schema
+// as their response content — no envelope.
+type LiteLLMExecutor struct {
+	baseURL    string
+	apiKey     string
+	httpClient *http.Client
+}
+
+// NewLiteLLM creates a LiteLLMExecutor.
+// timeout applies to the full HTTP round-trip per call.
+func NewLiteLLM(baseURL, apiKey string, timeout time.Duration) *LiteLLMExecutor {
+	return &LiteLLMExecutor{
+		baseURL: baseURL,
+		apiKey:  apiKey,
+		httpClient: &http.Client{Timeout: timeout},
+	}
+}
+
+type litellmMessage struct {
+	Role    string `json:"role"`
+	Content string `json:"content"`
+}
+
+type litellmRequest struct {
+	Model    string           `json:"model"`
+	Messages []litellmMessage `json:"messages"`
+}
+
+type litellmChoice struct {
+	Message litellmMessage `json:"message"`
+}
+
+type litellmResponse struct {
+	Choices []litellmChoice `json:"choices"`
+}
+
+// Run dispatches req to the LiteLLM server and parses the Result from the
+// assistant message content. Returns an error on network failure, non-200
+// status, or unparseable/invalid JSON — all of which the Orchestrator treats
+// as automatic escalation triggers.
+func (e *LiteLLMExecutor) Run(ctx context.Context, req Request) (Result, error) {
+	body := litellmRequest{
+		Model: req.Model,
+		Messages: []litellmMessage{
+			{Role: "system", Content: req.SkillPrompt},
+			{Role: "user", Content: req.TaskPrompt},
+		},
+	}
+
+	bodyBytes, err := json.Marshal(body)
+	if err != nil {
+		return Result{}, fmt.Errorf("litellm: marshal request: %w", err)
+	}
+
+	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, e.baseURL+"/v1/chat/completions", bytes.NewReader(bodyBytes))
+	if err != nil {
+		return Result{}, fmt.Errorf("litellm: create request: %w", err)
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+	if e.apiKey != "" {
+		httpReq.Header.Set("Authorization", "Bearer "+e.apiKey)
+	}
+
+	resp, err := e.httpClient.Do(httpReq)
+	if err != nil {
+		return Result{}, fmt.Errorf("litellm: request failed: %w", err)
+	}
+	defer resp.Body.Close() //nolint:errcheck
+
+	if resp.StatusCode != http.StatusOK {
+		return Result{}, fmt.Errorf("litellm: server returned status %d", resp.StatusCode)
+	}
+
+	var chatResp litellmResponse
+	if err := json.NewDecoder(resp.Body).Decode(&chatResp); err != nil {
+		return Result{}, fmt.Errorf("litellm: decode response: %w", err)
+	}
+	if len(chatResp.Choices) == 0 {
+		return Result{}, fmt.Errorf("litellm: no choices in response")
+	}
+
+	content := chatResp.Choices[0].Message.Content
+	var result Result
+	if err := json.Unmarshal([]byte(content), &result); err != nil {
+		return Result{}, fmt.Errorf("litellm: parse result JSON: %w — content: %s", err, content)
+	}
+	if err := result.Validate(); err != nil {
+		return Result{}, fmt.Errorf("litellm: invalid result: %w", err)
+	}
+	return result, nil
+}