diff options
Diffstat (limited to 'internal/llm/openai.go')
| -rw-r--r-- | internal/llm/openai.go | 282 |
1 files changed, 143 insertions, 139 deletions
diff --git a/internal/llm/openai.go b/internal/llm/openai.go index 288622f..fe6705b 100644 --- a/internal/llm/openai.go +++ b/internal/llm/openai.go @@ -3,17 +3,17 @@ package llm import ( - "bufio" - "bytes" - "context" - "encoding/json" - "errors" - "fmt" + "bufio" + "bytes" + "context" + "encoding/json" + "errors" + "fmt" "net/http" "strings" - "time" + "time" - "hexai/internal/logging" + "hexai/internal/logging" ) // openAIClient implements Client against OpenAI's Chat Completions API. @@ -22,10 +22,9 @@ type openAIClient struct { apiKey string baseURL string defaultModel string + chatLogger *logging.ChatLogger } -// Colors and base styling are provided by logging.go - // newOpenAI constructs an OpenAI client using explicit configuration values. // The apiKey may be empty; calls will fail until a valid key is supplied. func newOpenAI(baseURL, model, apiKey string) Client { @@ -40,16 +39,17 @@ func newOpenAI(baseURL, model, apiKey string) Client { apiKey: apiKey, baseURL: baseURL, defaultModel: model, + chatLogger: logging.NewChatLogger("openai"), } } type oaChatRequest struct { - Model string `json:"model"` - Messages []oaMessage `json:"messages"` - Temperature *float64 `json:"temperature,omitempty"` - MaxTokens *int `json:"max_tokens,omitempty"` - Stop []string `json:"stop,omitempty"` - Stream bool `json:"stream,omitempty"` + Model string `json:"model"` + Messages []oaMessage `json:"messages"` + Temperature *float64 `json:"temperature,omitempty"` + MaxTokens *int `json:"max_tokens,omitempty"` + Stop []string `json:"stop,omitempty"` + Stream bool `json:"stream,omitempty"` } type oaMessage struct { @@ -86,11 +86,18 @@ func (c *openAIClient) Chat(ctx context.Context, messages []Message, opts ...Req o.Model = c.defaultModel } start := time.Now() - logging.Logf("llm/openai ", "chat start model=%s temp=%.2f max_tokens=%d stop=%d messages=%d", o.Model, o.Temperature, o.MaxTokens, len(o.Stop), len(messages)) + logMessages := make([]struct { + Role string + Content string + }, len(messages)) for i, m := range messages { - // Sending context (cyan) - logging.Logf("llm/openai ", "msg[%d] role=%s size=%d preview=%s%s%s", i, m.Role, len(m.Content), logging.AnsiCyan, logging.PreviewForLog(m.Content), logging.AnsiBase) + logMessages[i] = struct { + Role string + Content string + }{Role: m.Role, Content: m.Content} } + c.chatLogger.LogStart(false, o.Model, o.Temperature, o.MaxTokens, o.Stop, logMessages) + req := oaChatRequest{Model: o.Model} req.Messages = make([]oaMessage, len(messages)) for i, m := range messages { @@ -152,138 +159,135 @@ func (c *openAIClient) Chat(ctx context.Context, messages []Message, opts ...Req return content, nil } -// small helper to keep return type consistent -func nilStringErr(msg string) (string, error) { return "", errors.New(msg) } - func (c *openAIClient) logf(format string, args ...any) { logging.Logf("llm/openai ", format, args...) } -func trimPreview(s string, n int) string { - if n <= 0 || len(s) <= n { - return s - } - return s[:n] + "…" -} - // Provider metadata func (c *openAIClient) Name() string { return "openai" } func (c *openAIClient) DefaultModel() string { return c.defaultModel } // Streaming support (optional) type oaStreamChunk struct { - Choices []struct { - Delta struct { - Content string `json:"content"` - } `json:"delta"` - FinishReason string `json:"finish_reason"` - } `json:"choices"` - Error *struct { - Message string `json:"message"` - Type string `json:"type"` - Param any `json:"param"` - Code any `json:"code"` - } `json:"error,omitempty"` + Choices []struct { + Delta struct { + Content string `json:"content"` + } `json:"delta"` + FinishReason string `json:"finish_reason"` + } `json:"choices"` + Error *struct { + Message string `json:"message"` + Type string `json:"type"` + Param any `json:"param"` + Code any `json:"code"` + } `json:"error,omitempty"` } func (c *openAIClient) ChatStream(ctx context.Context, messages []Message, onDelta func(string), opts ...RequestOption) error { - if c.apiKey == "" { - return errors.New("missing OpenAI API key") - } - o := Options{Model: c.defaultModel} - for _, opt := range opts { - opt(&o) - } - if o.Model == "" { - o.Model = c.defaultModel - } - start := time.Now() - logging.Logf("llm/openai ", "stream start model=%s temp=%.2f max_tokens=%d stop=%d messages=%d", o.Model, o.Temperature, o.MaxTokens, len(o.Stop), len(messages)) - for i, m := range messages { - logging.Logf("llm/openai ", "msg[%d] role=%s size=%d preview=%s%s%s", i, m.Role, len(m.Content), logging.AnsiCyan, logging.PreviewForLog(m.Content), logging.AnsiBase) - } + if c.apiKey == "" { + return errors.New("missing OpenAI API key") + } + o := Options{Model: c.defaultModel} + for _, opt := range opts { + opt(&o) + } + if o.Model == "" { + o.Model = c.defaultModel + } + start := time.Now() + logMessages := make([]struct { + Role string + Content string + }, len(messages)) + for i, m := range messages { + logMessages[i] = struct { + Role string + Content string + }{Role: m.Role, Content: m.Content} + } + c.chatLogger.LogStart(true, o.Model, o.Temperature, o.MaxTokens, o.Stop, logMessages) - req := oaChatRequest{Model: o.Model, Stream: true} - req.Messages = make([]oaMessage, len(messages)) - for i, m := range messages { - req.Messages[i] = oaMessage{Role: m.Role, Content: m.Content} - } - if o.Temperature != 0 { - req.Temperature = &o.Temperature - } - if o.MaxTokens > 0 { - req.MaxTokens = &o.MaxTokens - } - if len(o.Stop) > 0 { - req.Stop = o.Stop - } + req := oaChatRequest{Model: o.Model, Stream: true} + req.Messages = make([]oaMessage, len(messages)) + for i, m := range messages { + req.Messages[i] = oaMessage{Role: m.Role, Content: m.Content} + } + if o.Temperature != 0 { + req.Temperature = &o.Temperature + } + if o.MaxTokens > 0 { + req.MaxTokens = &o.MaxTokens + } + if len(o.Stop) > 0 { + req.Stop = o.Stop + } - body, err := json.Marshal(req) - if err != nil { - c.logf("marshal error: %v", err) - return err - } - endpoint := c.baseURL + "/chat/completions" - logging.Logf("llm/openai ", "POST %s (stream)", endpoint) - httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, endpoint, bytes.NewReader(body)) - if err != nil { - c.logf("new request error: %v", err) - return err - } - httpReq.Header.Set("Content-Type", "application/json") - httpReq.Header.Set("Authorization", "Bearer "+c.apiKey) - // Streaming uses SSE-style data lines - httpReq.Header.Set("Accept", "text/event-stream") + body, err := json.Marshal(req) + if err != nil { + c.logf("marshal error: %v", err) + return err + } + endpoint := c.baseURL + "/chat/completions" + logging.Logf("llm/openai ", "POST %s (stream)", endpoint) + httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, endpoint, bytes.NewReader(body)) + if err != nil { + c.logf("new request error: %v", err) + return err + } + httpReq.Header.Set("Content-Type", "application/json") + httpReq.Header.Set("Authorization", "Bearer "+c.apiKey) + // Streaming uses SSE-style data lines + httpReq.Header.Set("Accept", "text/event-stream") - resp, err := c.httpClient.Do(httpReq) - if err != nil { - logging.Logf("llm/openai ", "%shttp error after %s: %v%s", logging.AnsiRed, time.Since(start), err, logging.AnsiBase) - return err - } - defer resp.Body.Close() - if resp.StatusCode < 200 || resp.StatusCode >= 300 { - // try to decode body to surface message - var apiErr oaChatResponse - _ = json.NewDecoder(resp.Body).Decode(&apiErr) - if apiErr.Error != nil && apiErr.Error.Message != "" { - logging.Logf("llm/openai ", "%sapi error status=%d type=%s msg=%s duration=%s%s", logging.AnsiRed, resp.StatusCode, apiErr.Error.Type, apiErr.Error.Message, time.Since(start), logging.AnsiBase) - return fmt.Errorf("openai error: %s (status %d)", apiErr.Error.Message, resp.StatusCode) - } - logging.Logf("llm/openai ", "%shttp non-2xx status=%d duration=%s%s", logging.AnsiRed, resp.StatusCode, time.Since(start), logging.AnsiBase) - return fmt.Errorf("openai http error: status %d", resp.StatusCode) - } + resp, err := c.httpClient.Do(httpReq) + if err != nil { + logging.Logf("llm/openai ", "%shttp error after %s: %v%s", logging.AnsiRed, time.Since(start), err, logging.AnsiBase) + return err + } + defer resp.Body.Close() + if resp.StatusCode < 200 || resp.StatusCode >= 300 { + // try to decode body to surface message + var apiErr oaChatResponse + _ = json.NewDecoder(resp.Body).Decode(&apiErr) + if apiErr.Error != nil && apiErr.Error.Message != "" { + logging.Logf("llm/openai ", "%sapi error status=%d type=%s msg=%s duration=%s%s", logging.AnsiRed, resp.StatusCode, apiErr.Error.Type, apiErr.Error.Message, time.Since(start), logging.AnsiBase) + return fmt.Errorf("openai error: %s (status %d)", apiErr.Error.Message, resp.StatusCode) + } + logging.Logf("llm/openai ", "%shttp non-2xx status=%d duration=%s%s", logging.AnsiRed, resp.StatusCode, time.Since(start), logging.AnsiBase) + return fmt.Errorf("openai http error: status %d", resp.StatusCode) + } - // Parse SSE: lines starting with "data: " containing JSON or [DONE] - scanner := bufio.NewScanner(resp.Body) - // Increase buffer for long lines - const maxBuf = 1024 * 1024 - buf := make([]byte, 0, 64*1024) - scanner.Buffer(buf, maxBuf) - for scanner.Scan() { - line := scanner.Text() - if !strings.HasPrefix(line, "data: ") { - continue - } - payload := strings.TrimPrefix(line, "data: ") - if strings.TrimSpace(payload) == "[DONE]" { - break - } - var chunk oaStreamChunk - if err := json.Unmarshal([]byte(payload), &chunk); err != nil { - continue // skip malformed lines - } - if chunk.Error != nil && chunk.Error.Message != "" { - logging.Logf("llm/openai ", "%sstream error: %s%s", logging.AnsiRed, chunk.Error.Message, logging.AnsiBase) - return fmt.Errorf("openai stream error: %s", chunk.Error.Message) - } - for _, ch := range chunk.Choices { - if ch.Delta.Content != "" { - onDelta(ch.Delta.Content) - } - } - } - if err := scanner.Err(); err != nil { - logging.Logf("llm/openai ", "%sstream read error after %s: %v%s", logging.AnsiRed, time.Since(start), err, logging.AnsiBase) - return err - } - logging.Logf("llm/openai ", "stream end duration=%s", time.Since(start)) - return nil -} + // Parse SSE: lines starting with "data: " containing JSON or [DONE] + scanner := bufio.NewScanner(resp.Body) + // Increase buffer for long lines + const maxBuf = 1024 * 1024 + buf := make([]byte, 0, 64*1024) + scanner.Buffer(buf, maxBuf) + for scanner.Scan() { + line := scanner.Text() + if !strings.HasPrefix(line, "data: ") { + continue + } + payload := strings.TrimPrefix(line, "data: ") + if strings.TrimSpace(payload) == "[DONE]" { + break + } + var chunk oaStreamChunk + if err := json.Unmarshal([]byte(payload), &chunk); err != nil { + continue // skip malformed lines + } + if chunk.Error != nil && chunk.Error.Message != "" { + logging.Logf("llm/openai ", "%sstream error: %s%s", logging.AnsiRed, chunk.Error.Message, logging.AnsiBase) + return fmt.Errorf("openai stream error: %s", chunk.Error.Message) + } + for _, ch := range chunk.Choices { + if ch.Delta.Content != "" { + onDelta(ch.Delta.Content) + } + } + } + if err := scanner.Err(); err != nil { + logging.Logf("llm/openai ", "%sstream read error after %s: %v%s", logging.AnsiRed, time.Since(start), err, logging.AnsiBase) + return err + } + logging.Logf("llm/openai ", "stream end duration=%s", time.Since(start)) + return nil +}
\ No newline at end of file |