provider.go 8.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338
  1. // Package hyper provides a fantasy.Provider that proxies requests to Hyper.
  2. package hyper
  3. import (
  4. "bufio"
  5. "bytes"
  6. "cmp"
  7. "context"
  8. _ "embed"
  9. "encoding/json"
  10. "errors"
  11. "fmt"
  12. "io"
  13. "log/slog"
  14. "maps"
  15. "net/http"
  16. "net/url"
  17. "os"
  18. "strconv"
  19. "strings"
  20. "sync"
  21. "time"
  22. "charm.land/catwalk/pkg/catwalk"
  23. "charm.land/fantasy"
  24. "charm.land/fantasy/object"
  25. "github.com/charmbracelet/crush/internal/event"
  26. )
  27. //go:generate wget -O provider.json https://hyper.charm.land/api/v1/provider
  28. //go:embed provider.json
  29. var embedded []byte
  30. // Enabled returns true if hyper is enabled.
  31. var Enabled = sync.OnceValue(func() bool {
  32. b, _ := strconv.ParseBool(
  33. cmp.Or(
  34. os.Getenv("HYPER"),
  35. os.Getenv("HYPERCRUSH"),
  36. os.Getenv("HYPER_ENABLE"),
  37. os.Getenv("HYPER_ENABLED"),
  38. ),
  39. )
  40. return b
  41. })
  42. // Embedded returns the embedded Hyper provider.
  43. var Embedded = sync.OnceValue(func() catwalk.Provider {
  44. var provider catwalk.Provider
  45. if err := json.Unmarshal(embedded, &provider); err != nil {
  46. slog.Error("Could not use embedded provider data", "err", err)
  47. }
  48. if e := os.Getenv("HYPER_URL"); e != "" {
  49. provider.APIEndpoint = e + "/api/v1/fantasy"
  50. }
  51. return provider
  52. })
  53. const (
  54. // Name is the default name of this meta provider.
  55. Name = "hyper"
  56. // defaultBaseURL is the default proxy URL.
  57. defaultBaseURL = "https://hyper.charm.land"
  58. )
  59. // BaseURL returns the base URL, which is either $HYPER_URL or the default.
  60. var BaseURL = sync.OnceValue(func() string {
  61. return cmp.Or(os.Getenv("HYPER_URL"), defaultBaseURL)
  62. })
  63. var ErrNoCredits = errors.New("you're out of credits")
  64. type options struct {
  65. baseURL string
  66. apiKey string
  67. name string
  68. headers map[string]string
  69. client *http.Client
  70. }
  71. // Option configures the proxy provider.
  72. type Option = func(*options)
  73. // New creates a new proxy provider.
  74. func New(opts ...Option) (fantasy.Provider, error) {
  75. o := options{
  76. baseURL: BaseURL() + "/api/v1/fantasy",
  77. name: Name,
  78. headers: map[string]string{
  79. "x-crush-id": event.GetID(),
  80. },
  81. client: &http.Client{Timeout: 0}, // stream-safe
  82. }
  83. for _, opt := range opts {
  84. opt(&o)
  85. }
  86. return &provider{options: o}, nil
  87. }
  88. // WithBaseURL sets the proxy base URL (e.g. http://localhost:8080).
  89. func WithBaseURL(url string) Option { return func(o *options) { o.baseURL = url } }
  90. // WithName sets the provider name.
  91. func WithName(name string) Option { return func(o *options) { o.name = name } }
  92. // WithHeaders sets extra headers sent to the proxy.
  93. func WithHeaders(headers map[string]string) Option {
  94. return func(o *options) {
  95. maps.Copy(o.headers, headers)
  96. }
  97. }
  98. // WithHTTPClient sets custom HTTP client.
  99. func WithHTTPClient(c *http.Client) Option { return func(o *options) { o.client = c } }
  100. // WithAPIKey sets the API key.
  101. func WithAPIKey(key string) Option {
  102. return func(o *options) {
  103. o.apiKey = key
  104. }
  105. }
  106. type provider struct{ options options }
  107. func (p *provider) Name() string { return p.options.name }
  108. // LanguageModel implements fantasy.Provider.
  109. func (p *provider) LanguageModel(_ context.Context, modelID string) (fantasy.LanguageModel, error) {
  110. if modelID == "" {
  111. return nil, errors.New("missing model id")
  112. }
  113. return &languageModel{modelID: modelID, provider: p.options.name, opts: p.options}, nil
  114. }
  115. type languageModel struct {
  116. provider string
  117. modelID string
  118. opts options
  119. }
  120. // GenerateObject implements fantasy.LanguageModel.
  121. func (m *languageModel) GenerateObject(ctx context.Context, call fantasy.ObjectCall) (*fantasy.ObjectResponse, error) {
  122. return object.GenerateWithTool(ctx, m, call)
  123. }
  124. // StreamObject implements fantasy.LanguageModel.
  125. func (m *languageModel) StreamObject(ctx context.Context, call fantasy.ObjectCall) (fantasy.ObjectStreamResponse, error) {
  126. return object.StreamWithTool(ctx, m, call)
  127. }
  128. func (m *languageModel) Provider() string { return m.provider }
  129. func (m *languageModel) Model() string { return m.modelID }
  130. // Generate implements fantasy.LanguageModel by calling the proxy JSON endpoint.
  131. func (m *languageModel) Generate(ctx context.Context, call fantasy.Call) (*fantasy.Response, error) {
  132. resp, err := m.doRequest(ctx, false, call)
  133. if err != nil {
  134. return nil, err
  135. }
  136. defer func() { _ = resp.Body.Close() }()
  137. if resp.StatusCode < 200 || resp.StatusCode >= 300 {
  138. b, _ := ioReadAllLimit(resp.Body, 64*1024)
  139. return nil, fmt.Errorf("proxy generate error: %s", strings.TrimSpace(string(b)))
  140. }
  141. var out fantasy.Response
  142. if err := json.NewDecoder(resp.Body).Decode(&out); err != nil {
  143. return nil, err
  144. }
  145. return &out, nil
  146. }
  147. // Stream implements fantasy.LanguageModel using SSE from the proxy.
  148. func (m *languageModel) Stream(ctx context.Context, call fantasy.Call) (fantasy.StreamResponse, error) {
  149. // Prefer explicit /stream endpoint
  150. resp, err := m.doRequest(ctx, true, call)
  151. if err != nil {
  152. return nil, err
  153. }
  154. switch resp.StatusCode {
  155. case http.StatusTooManyRequests:
  156. _ = resp.Body.Close()
  157. return nil, toProviderError(resp, retryAfter(resp))
  158. case http.StatusUnauthorized:
  159. _ = resp.Body.Close()
  160. return nil, toProviderError(resp, "")
  161. case http.StatusPaymentRequired:
  162. _ = resp.Body.Close()
  163. return nil, ErrNoCredits
  164. }
  165. if resp.StatusCode < 200 || resp.StatusCode >= 300 {
  166. defer func() { _ = resp.Body.Close() }()
  167. b, _ := ioReadAllLimit(resp.Body, 64*1024)
  168. return nil, &fantasy.ProviderError{
  169. Title: "Stream Error",
  170. Message: strings.TrimSpace(string(b)),
  171. StatusCode: resp.StatusCode,
  172. }
  173. }
  174. return func(yield func(fantasy.StreamPart) bool) {
  175. defer func() { _ = resp.Body.Close() }()
  176. scanner := bufio.NewScanner(resp.Body)
  177. buf := make([]byte, 0, 64*1024)
  178. scanner.Buffer(buf, 4*1024*1024)
  179. var (
  180. event string
  181. dataBuf bytes.Buffer
  182. sawFinish bool
  183. dispatch = func() bool {
  184. if dataBuf.Len() == 0 || event == "" {
  185. dataBuf.Reset()
  186. event = ""
  187. return true
  188. }
  189. var part fantasy.StreamPart
  190. if err := json.Unmarshal(dataBuf.Bytes(), &part); err != nil {
  191. return yield(fantasy.StreamPart{Type: fantasy.StreamPartTypeError, Error: err})
  192. }
  193. if part.Type == fantasy.StreamPartTypeFinish {
  194. sawFinish = true
  195. }
  196. ok := yield(part)
  197. dataBuf.Reset()
  198. event = ""
  199. return ok
  200. }
  201. )
  202. for scanner.Scan() {
  203. line := scanner.Text()
  204. if line == "" { // event boundary
  205. if !dispatch() {
  206. return
  207. }
  208. continue
  209. }
  210. if strings.HasPrefix(line, ":") { // comment / ping
  211. continue
  212. }
  213. if strings.HasPrefix(line, "event: ") {
  214. event = strings.TrimSpace(line[len("event: "):])
  215. continue
  216. }
  217. if strings.HasPrefix(line, "data: ") {
  218. if dataBuf.Len() > 0 {
  219. dataBuf.WriteByte('\n')
  220. }
  221. dataBuf.WriteString(line[len("data: "):])
  222. continue
  223. }
  224. }
  225. if err := scanner.Err(); err != nil {
  226. if sawFinish && (errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded)) {
  227. // If we already saw an explicit finish event, treat cancellation as a no-op.
  228. } else {
  229. _ = yield(fantasy.StreamPart{Type: fantasy.StreamPartTypeError, Error: err})
  230. return
  231. }
  232. }
  233. if err := ctx.Err(); err != nil && !sawFinish {
  234. _ = yield(fantasy.StreamPart{Type: fantasy.StreamPartTypeError, Error: err})
  235. return
  236. }
  237. // flush any pending data
  238. _ = dispatch()
  239. if !sawFinish {
  240. _ = yield(fantasy.StreamPart{Type: fantasy.StreamPartTypeFinish})
  241. }
  242. }, nil
  243. }
  244. func (m *languageModel) doRequest(ctx context.Context, stream bool, call fantasy.Call) (*http.Response, error) {
  245. addr, err := url.Parse(m.opts.baseURL)
  246. if err != nil {
  247. return nil, err
  248. }
  249. addr = addr.JoinPath(m.modelID)
  250. if stream {
  251. addr = addr.JoinPath("stream")
  252. } else {
  253. addr = addr.JoinPath("generate")
  254. }
  255. body, err := json.Marshal(call)
  256. if err != nil {
  257. return nil, err
  258. }
  259. req, err := http.NewRequestWithContext(ctx, http.MethodPost, addr.String(), bytes.NewReader(body))
  260. if err != nil {
  261. return nil, err
  262. }
  263. req.Header.Set("Content-Type", "application/json")
  264. if stream {
  265. req.Header.Set("Accept", "text/event-stream")
  266. } else {
  267. req.Header.Set("Accept", "application/json")
  268. }
  269. for k, v := range m.opts.headers {
  270. req.Header.Set(k, v)
  271. }
  272. if m.opts.apiKey != "" {
  273. req.Header.Set("Authorization", m.opts.apiKey)
  274. }
  275. return m.opts.client.Do(req)
  276. }
  277. // ioReadAllLimit reads up to n bytes.
  278. func ioReadAllLimit(r io.Reader, n int64) ([]byte, error) {
  279. var b bytes.Buffer
  280. if n <= 0 {
  281. n = 1 << 20
  282. }
  283. lr := &io.LimitedReader{R: r, N: n}
  284. _, err := b.ReadFrom(lr)
  285. return b.Bytes(), err
  286. }
  287. func toProviderError(resp *http.Response, message string) error {
  288. return &fantasy.ProviderError{
  289. Title: fantasy.ErrorTitleForStatusCode(resp.StatusCode),
  290. Message: message,
  291. StatusCode: resp.StatusCode,
  292. }
  293. }
  294. func retryAfter(resp *http.Response) string {
  295. after, err := strconv.Atoi(resp.Header.Get("Retry-After"))
  296. if err == nil && after > 0 {
  297. d := time.Duration(after) * time.Second
  298. return "Try again in " + d.String()
  299. }
  300. return "Try again later"
  301. }