modeldispatcher-browser-agent 0.6.0 → 0.6.6

package/src/providers/ollama.ts

@@ -1,81 +1,81 @@
-/**
- * Ollama — a local (or self-hosted) model server, no API key. Always uses
- * `/api/chat` (message-list based) rather than switching between `/generate`
- * and `/chat` per call shape — a single-turn prompt is just a one-message
- * chat, so one code path covers both without the caller needing to know
- * which Ollama endpoint that implies.
- */
-
-import { consumeLines, fetchWithRetry, validateOllamaUrl } from "../http.js";
-import type { ChatMessage, ProviderCallConfig, RequestOptions } from "../types.js";
-
-const DEFAULT_MAX_TOKENS = 1024;
-
-function toApiMessages(messages: readonly ChatMessage[], systemInstruction?: string) {
-  const hasSystemTurn = messages.some((m) => m.role === "system");
-  const system = !hasSystemTurn && systemInstruction ? [{ role: "system", content: systemInstruction }] : [];
-  return [...system, ...messages.map((m) => ({ role: m.role, content: m.content }))];
-}
-
-function buildBody(cfg: ProviderCallConfig, messages: readonly ChatMessage[], options: RequestOptions, stream: boolean) {
-  return {
-    model: cfg.model,
-    messages: toApiMessages(messages, options.systemInstruction),
-    stream,
-    options: { num_predict: options.maxTokens ?? DEFAULT_MAX_TOKENS },
-  };
-}
-
-export async function completeOllama(
-  cfg: ProviderCallConfig,
-  messages: readonly ChatMessage[],
-  options: RequestOptions = {},
-): Promise<string> {
-  const baseUrl = validateOllamaUrl(cfg.ollamaUrl);
-  const res = await fetchWithRetry(
-    `${baseUrl}/api/chat`,
-    {
-      method: "POST",
-      headers: { "Content-Type": "application/json" },
-      body: JSON.stringify(buildBody(cfg, messages, options, false)),
-    },
-    options.signal,
-  );
-  if (!res.ok) throw new Error(`Ollama error: HTTP ${res.status}. Is Ollama running?`);
-  const data = await res.json();
-  return data.message?.content ?? "";
-}
-
-export async function streamOllama(
-  cfg: ProviderCallConfig,
-  messages: readonly ChatMessage[],
-  onChunk: (fullTextSoFar: string) => void,
-  options: RequestOptions = {},
-): Promise<string> {
-  const baseUrl = validateOllamaUrl(cfg.ollamaUrl);
-  const res = await fetchWithRetry(
-    `${baseUrl}/api/chat`,
-    {
-      method: "POST",
-      headers: { "Content-Type": "application/json" },
-      body: JSON.stringify(buildBody(cfg, messages, options, true)),
-    },
-    options.signal,
-  );
-  if (!res.ok) throw new Error(`Ollama error: HTTP ${res.status}. Is Ollama running?`);
-
-  let full = "";
-  await consumeLines(res, (line) => {
-    if (!line.trim()) return;
-    try {
-      const text = JSON.parse(line).message?.content;
-      if (text) {
-        full += text;
-        onChunk(full);
-      }
-    } catch {
-      // Skip a malformed/partial NDJSON line.
-    }
-  });
-  return full;
-}
+/**
+ * Ollama — a local (or self-hosted) model server, no API key. Always uses
+ * `/api/chat` (message-list based) rather than switching between `/generate`
+ * and `/chat` per call shape — a single-turn prompt is just a one-message
+ * chat, so one code path covers both without the caller needing to know
+ * which Ollama endpoint that implies.
+ */
+
+import { consumeLines, fetchWithRetry, validateOllamaUrl } from "../http.js";
+import type { ChatMessage, ProviderCallConfig, RequestOptions } from "../types.js";
+
+const DEFAULT_MAX_TOKENS = 1024;
+
+function toApiMessages(messages: readonly ChatMessage[], systemInstruction?: string) {
+  const hasSystemTurn = messages.some((m) => m.role === "system");
+  const system = !hasSystemTurn && systemInstruction ? [{ role: "system", content: systemInstruction }] : [];
+  return [...system, ...messages.map((m) => ({ role: m.role, content: m.content }))];
+}
+
+function buildBody(cfg: ProviderCallConfig, messages: readonly ChatMessage[], options: RequestOptions, stream: boolean) {
+  return {
+    model: cfg.model,
+    messages: toApiMessages(messages, options.systemInstruction),
+    stream,
+    options: { num_predict: options.maxTokens ?? DEFAULT_MAX_TOKENS },
+  };
+}
+
+export async function completeOllama(
+  cfg: ProviderCallConfig,
+  messages: readonly ChatMessage[],
+  options: RequestOptions = {},
+): Promise<string> {
+  const baseUrl = validateOllamaUrl(cfg.ollamaUrl);
+  const res = await fetchWithRetry(
+    `${baseUrl}/api/chat`,
+    {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify(buildBody(cfg, messages, options, false)),
+    },
+    options.signal,
+  );
+  if (!res.ok) throw new Error(`Ollama error: HTTP ${res.status}. Is Ollama running?`);
+  const data = await res.json();
+  return data.message?.content ?? "";
+}
+
+export async function streamOllama(
+  cfg: ProviderCallConfig,
+  messages: readonly ChatMessage[],
+  onChunk: (fullTextSoFar: string) => void,
+  options: RequestOptions = {},
+): Promise<string> {
+  const baseUrl = validateOllamaUrl(cfg.ollamaUrl);
+  const res = await fetchWithRetry(
+    `${baseUrl}/api/chat`,
+    {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify(buildBody(cfg, messages, options, true)),
+    },
+    options.signal,
+  );
+  if (!res.ok) throw new Error(`Ollama error: HTTP ${res.status}. Is Ollama running?`);
+
+  let full = "";
+  await consumeLines(res, (line) => {
+    if (!line.trim()) return;
+    try {
+      const text = JSON.parse(line).message?.content;
+      if (text) {
+        full += text;
+        onChunk(full);
+      }
+    } catch {
+      // Skip a malformed/partial NDJSON line.
+    }
+  });
+  return full;
+}

package/src/providers/openaiCompatible.ts

@@ -1,89 +1,89 @@
-/**
- * OpenAI and Groq — both speak the same chat-completions REST shape, direct
- * from the browser, no SDK. Groq is just the same code pointed at its own
- * base URL (matches how the Python core's GroqProvider relates to
- * OpenAIProvider — same idea, browser side).
- */
-
-import { consumeLines, describeHttpError, fetchWithRetry } from "../http.js";
-import type { ChatMessage, ProviderCallConfig, RequestOptions } from "../types.js";
-
-const DEFAULT_MAX_TOKENS = 1024;
-
-const BASE_URLS: Record<"openai" | "groq", string> = {
-  openai: "https://api.openai.com/v1/chat/completions",
-  groq: "https://api.groq.com/openai/v1/chat/completions",
-};
-
-function toApiMessages(messages: readonly ChatMessage[], systemInstruction?: string) {
-  const hasSystemTurn = messages.some((m) => m.role === "system");
-  const system = !hasSystemTurn && systemInstruction ? [{ role: "system", content: systemInstruction }] : [];
-  return [...system, ...messages.map((m) => ({ role: m.role, content: m.content }))];
-}
-
-function buildBody(cfg: ProviderCallConfig, messages: readonly ChatMessage[], options: RequestOptions, stream: boolean) {
-  return {
-    model: cfg.model,
-    max_tokens: options.maxTokens ?? DEFAULT_MAX_TOKENS,
-    messages: toApiMessages(messages, options.systemInstruction),
-    ...(options.jsonMode ? { response_format: { type: "json_object" } } : {}),
-    ...(stream ? { stream: true } : {}),
-  };
-}
-
-async function request(
-  provider: "openai" | "groq",
-  cfg: ProviderCallConfig,
-  messages: readonly ChatMessage[],
-  options: RequestOptions,
-  stream: boolean,
-): Promise<Response> {
-  const res = await fetchWithRetry(
-    BASE_URLS[provider],
-    {
-      method: "POST",
-      headers: { "Content-Type": "application/json", Authorization: `Bearer ${cfg.apiKey}` },
-      body: JSON.stringify(buildBody(cfg, messages, options, stream)),
-    },
-    options.signal,
-  );
-  if (!res.ok) throw new Error(await describeHttpError(res));
-  return res;
-}
-
-export async function completeOpenAICompatible(
-  provider: "openai" | "groq",
-  cfg: ProviderCallConfig,
-  messages: readonly ChatMessage[],
-  options: RequestOptions = {},
-): Promise<string> {
-  const res = await request(provider, cfg, messages, options, false);
-  const data = await res.json();
-  return data.choices?.[0]?.message?.content ?? "";
-}
-
-export async function streamOpenAICompatible(
-  provider: "openai" | "groq",
-  cfg: ProviderCallConfig,
-  messages: readonly ChatMessage[],
-  onChunk: (fullTextSoFar: string) => void,
-  options: RequestOptions = {},
-): Promise<string> {
-  const res = await request(provider, cfg, messages, options, true);
-  let full = "";
-  await consumeLines(res, (line) => {
-    if (!line.startsWith("data: ")) return;
-    const raw = line.slice(6).trim();
-    if (raw === "[DONE]") return;
-    try {
-      const delta = JSON.parse(raw).choices?.[0]?.delta?.content;
-      if (delta) {
-        full += delta;
-        onChunk(full);
-      }
-    } catch {
-      // Skip a malformed/partial SSE frame.
-    }
-  });
-  return full;
-}
+/**
+ * OpenAI and Groq — both speak the same chat-completions REST shape, direct
+ * from the browser, no SDK. Groq is just the same code pointed at its own
+ * base URL (matches how the Python core's GroqProvider relates to
+ * OpenAIProvider — same idea, browser side).
+ */
+
+import { consumeLines, describeHttpError, fetchWithRetry } from "../http.js";
+import type { ChatMessage, ProviderCallConfig, RequestOptions } from "../types.js";
+
+const DEFAULT_MAX_TOKENS = 1024;
+
+const BASE_URLS: Record<"openai" | "groq", string> = {
+  openai: "https://api.openai.com/v1/chat/completions",
+  groq: "https://api.groq.com/openai/v1/chat/completions",
+};
+
+function toApiMessages(messages: readonly ChatMessage[], systemInstruction?: string) {
+  const hasSystemTurn = messages.some((m) => m.role === "system");
+  const system = !hasSystemTurn && systemInstruction ? [{ role: "system", content: systemInstruction }] : [];
+  return [...system, ...messages.map((m) => ({ role: m.role, content: m.content }))];
+}
+
+function buildBody(cfg: ProviderCallConfig, messages: readonly ChatMessage[], options: RequestOptions, stream: boolean) {
+  return {
+    model: cfg.model,
+    max_tokens: options.maxTokens ?? DEFAULT_MAX_TOKENS,
+    messages: toApiMessages(messages, options.systemInstruction),
+    ...(options.jsonMode ? { response_format: { type: "json_object" } } : {}),
+    ...(stream ? { stream: true } : {}),
+  };
+}
+
+async function request(
+  provider: "openai" | "groq",
+  cfg: ProviderCallConfig,
+  messages: readonly ChatMessage[],
+  options: RequestOptions,
+  stream: boolean,
+): Promise<Response> {
+  const res = await fetchWithRetry(
+    BASE_URLS[provider],
+    {
+      method: "POST",
+      headers: { "Content-Type": "application/json", Authorization: `Bearer ${cfg.apiKey}` },
+      body: JSON.stringify(buildBody(cfg, messages, options, stream)),
+    },
+    options.signal,
+  );
+  if (!res.ok) throw new Error(await describeHttpError(res));
+  return res;
+}
+
+export async function completeOpenAICompatible(
+  provider: "openai" | "groq",
+  cfg: ProviderCallConfig,
+  messages: readonly ChatMessage[],
+  options: RequestOptions = {},
+): Promise<string> {
+  const res = await request(provider, cfg, messages, options, false);
+  const data = await res.json();
+  return data.choices?.[0]?.message?.content ?? "";
+}
+
+export async function streamOpenAICompatible(
+  provider: "openai" | "groq",
+  cfg: ProviderCallConfig,
+  messages: readonly ChatMessage[],
+  onChunk: (fullTextSoFar: string) => void,
+  options: RequestOptions = {},
+): Promise<string> {
+  const res = await request(provider, cfg, messages, options, true);
+  let full = "";
+  await consumeLines(res, (line) => {
+    if (!line.startsWith("data: ")) return;
+    const raw = line.slice(6).trim();
+    if (raw === "[DONE]") return;
+    try {
+      const delta = JSON.parse(raw).choices?.[0]?.delta?.content;
+      if (delta) {
+        full += delta;
+        onChunk(full);
+      }
+    } catch {
+      // Skip a malformed/partial SSE frame.
+    }
+  });
+  return full;
+}

package/src/registry.ts

@@ -1,82 +1,82 @@
-/**
- * The provider registry — one canonical source for the metadata every app's
- * settings screen was independently duplicating (name, default model, key
- * placeholder, where to get a key).
- */
-
-import type { ProviderId, ProviderInfo } from "./types.js";
-
-export const PROVIDERS: Record<ProviderId, ProviderInfo> = {
-  gemini: {
-    id: "gemini",
-    name: "Google Gemini",
-    free: false,
-    defaultModel: "gemini-2.0-flash",
-    placeholder: "AIza...",
-    infoUrl: "https://aistudio.google.com/app/apikey",
-    infoText: "Get a free key from Google AI Studio",
-  },
-  groq: {
-    id: "groq",
-    name: "Groq",
-    free: true,
-    defaultModel: "llama-3.1-8b-instant",
-    placeholder: "gsk_...",
-    infoUrl: "https://console.groq.com/keys",
-    infoText: "Get a free key from Groq Console",
-  },
-  ollama: {
-    id: "ollama",
-    name: "Ollama (local)",
-    free: true,
-    noKey: true,
-    defaultModel: "llama3.2",
-    placeholder: "http://localhost:11434",
-    infoUrl: "https://ollama.ai",
-    infoText: "Install Ollama on your machine",
-  },
-  anthropic: {
-    id: "anthropic",
-    name: "Anthropic Claude",
-    free: false,
-    defaultModel: "claude-haiku-4-5-20251001",
-    placeholder: "sk-ant-...",
-    infoUrl: "https://console.anthropic.com/settings/keys",
-    infoText: "Get a key from the Anthropic Console",
-  },
-  openai: {
-    id: "openai",
-    name: "OpenAI",
-    free: false,
-    defaultModel: "gpt-4o-mini",
-    placeholder: "sk-...",
-    infoUrl: "https://platform.openai.com/api-keys",
-    infoText: "Get a key from the OpenAI Platform",
-  },
-};
-
-export function isKnownProvider(id: string): id is ProviderId {
-  return Object.hasOwn(PROVIDERS, id);
-}
-
-/**
- * A curated, hand-maintained shortlist of current models per vendor, for
- * rendering a model `<select>` instead of a free-text field. Necessarily
- * goes stale as vendors ship new models — update this list by hand when
- * that happens; it isn't fetched from anywhere. `PROVIDERS[id].defaultModel`
- * is always included (first) even if it's ever edited out of this list.
- */
-export const MODEL_OPTIONS: Record<ProviderId, readonly string[]> = {
-  gemini: ["gemini-2.0-flash", "gemini-2.5-flash", "gemini-2.5-pro"],
-  groq: ["llama-3.1-8b-instant", "llama-3.3-70b-versatile", "openai/gpt-oss-120b"],
-  ollama: ["llama3.2", "llama3.1", "mistral", "qwen2.5"],
-  anthropic: ["claude-haiku-4-5-20251001", "claude-sonnet-4-5", "claude-opus-4-8"],
-  openai: ["gpt-4o-mini", "gpt-4o", "gpt-4.1-mini"],
-};
-
-/** `MODEL_OPTIONS[provider]` with `defaultModel` guaranteed present and first. */
-export function modelOptionsFor(provider: ProviderId): readonly string[] {
-  const options = MODEL_OPTIONS[provider];
-  const defaultModel = PROVIDERS[provider].defaultModel;
-  return options.includes(defaultModel) ? options : [defaultModel, ...options];
-}
+/**
+ * The provider registry — one canonical source for the metadata every app's
+ * settings screen was independently duplicating (name, default model, key
+ * placeholder, where to get a key).
+ */
+
+import type { ProviderId, ProviderInfo } from "./types.js";
+
+export const PROVIDERS: Record<ProviderId, ProviderInfo> = {
+  gemini: {
+    id: "gemini",
+    name: "Google Gemini",
+    free: false,
+    defaultModel: "gemini-2.0-flash",
+    placeholder: "AIza...",
+    infoUrl: "https://aistudio.google.com/app/apikey",
+    infoText: "Get a free key from Google AI Studio",
+  },
+  groq: {
+    id: "groq",
+    name: "Groq",
+    free: true,
+    defaultModel: "llama-3.1-8b-instant",
+    placeholder: "gsk_...",
+    infoUrl: "https://console.groq.com/keys",
+    infoText: "Get a free key from Groq Console",
+  },
+  ollama: {
+    id: "ollama",
+    name: "Ollama (local)",
+    free: true,
+    noKey: true,
+    defaultModel: "llama3.2",
+    placeholder: "http://localhost:11434",
+    infoUrl: "https://ollama.ai",
+    infoText: "Install Ollama on your machine",
+  },
+  anthropic: {
+    id: "anthropic",
+    name: "Anthropic Claude",
+    free: false,
+    defaultModel: "claude-haiku-4-5-20251001",
+    placeholder: "sk-ant-...",
+    infoUrl: "https://console.anthropic.com/settings/keys",
+    infoText: "Get a key from the Anthropic Console",
+  },
+  openai: {
+    id: "openai",
+    name: "OpenAI",
+    free: false,
+    defaultModel: "gpt-4o-mini",
+    placeholder: "sk-...",
+    infoUrl: "https://platform.openai.com/api-keys",
+    infoText: "Get a key from the OpenAI Platform",
+  },
+};
+
+export function isKnownProvider(id: string): id is ProviderId {
+  return Object.hasOwn(PROVIDERS, id);
+}
+
+/**
+ * A curated, hand-maintained shortlist of current models per vendor, for
+ * rendering a model `<select>` instead of a free-text field. Necessarily
+ * goes stale as vendors ship new models — update this list by hand when
+ * that happens; it isn't fetched from anywhere. `PROVIDERS[id].defaultModel`
+ * is always included (first) even if it's ever edited out of this list.
+ */
+export const MODEL_OPTIONS: Record<ProviderId, readonly string[]> = {
+  gemini: ["gemini-2.0-flash", "gemini-2.5-flash", "gemini-2.5-pro"],
+  groq: ["llama-3.1-8b-instant", "llama-3.3-70b-versatile", "openai/gpt-oss-120b"],
+  ollama: ["llama3.2", "llama3.1", "mistral", "qwen2.5"],
+  anthropic: ["claude-haiku-4-5-20251001", "claude-sonnet-4-5", "claude-opus-4-8"],
+  openai: ["gpt-4o-mini", "gpt-4o", "gpt-4.1-mini"],
+};
+
+/** `MODEL_OPTIONS[provider]` with `defaultModel` guaranteed present and first. */
+export function modelOptionsFor(provider: ProviderId): readonly string[] {
+  const options = MODEL_OPTIONS[provider];
+  const defaultModel = PROVIDERS[provider].defaultModel;
+  return options.includes(defaultModel) ? options : [defaultModel, ...options];
+}