chainlesschain 0.166.108 → 0.166.112

package/src/commands/ask.js

@@ -29,7 +29,16 @@ import { captureAgentEvolutionRuntimeComposition } from "../lib/evolution/agent-
 import { captureAgentEvolutionIngress } from "../lib/evolution/agent-evolution-ingress.js";
 import { logger } from "../lib/logger.js";
 import { BUILT_IN_PROVIDERS } from "../lib/llm-providers.js";
+import { normalizeOllamaBaseUrl } from "../lib/ollama-base-url.js";
+import {
+  geminiGenerateContentUrl,
+  geminiRequestBody,
+  geminiResponseText,
+  normalizeProviderBaseUrl,
+  openAICompatibleBaseUrl,
+} from "../lib/provider-protocol.js";
 import { loadConfig } from "../lib/config-manager.js";
+import { resolveLlmConnection } from "../lib/llm-connection-resolution.js";
 import { expandFileRefs } from "../runtime/file-ref-expander.js";
 
 const ANDROID_LOCAL_OLLAMA_URL = "http://127.0.0.1:18484";
@@ -73,6 +82,33 @@ export function resolveOllamaBaseUrl({
   return "http://localhost:11434";
 }
 
+export function resolveAskOptions(
+  options = {},
+  config = {},
+  env = process.env,
+) {
+  const connection = resolveLlmConnection(options, [config.llm || {}], { env });
+  const provider = connection.provider;
+  const resolved = {
+    model: connection.model || (provider === "ollama" ? "qwen2:7b" : null),
+    provider,
+    apiKey: connection.apiKey,
+    baseUrl: connection.baseUrl,
+  };
+  if (provider === "ollama") {
+    resolved.baseUrl = resolveOllamaBaseUrl({
+      options,
+      env,
+      config:
+        config.llm?.provider === "ollama" &&
+        connection.endpointSources.includes(config.llm)
+          ? config
+          : {},
+    });
+  }
+  return resolved;
+}
+
 /**
  * Send a single question to an LLM provider.
  * @param {string} question
@@ -115,7 +151,8 @@ export async function queryLLM(
     throw error;
   }
   const provider = options.provider || "ollama";
-  const model = options.model || "qwen2:7b";
+  const model =
+    options.model || BUILT_IN_PROVIDERS[provider]?.models?.[0] || "qwen2:7b";
   const messages = [{ role: "user", content: question }];
   const providerMessages = (
     await captureAgentEvolutionIngress(evolutionIngress).prepareModelRequest({
@@ -139,7 +176,7 @@ export async function queryLLM(
   }
 
   if (provider === "ollama") {
-    const baseUrl = options.baseUrl || "http://localhost:11434";
+    const baseUrl = normalizeOllamaBaseUrl(options.baseUrl);
     // Use /api/chat (modern Ollama API + only endpoint Android LocalLlmServer
     // implements). Non-streaming mode keeps the response shape simple.
     const response = await providerFetch(`${baseUrl}/api/chat`, {
@@ -167,7 +204,6 @@ export async function queryLLM(
     return data.message?.content ?? "";
   }
 
-  // OpenAI-compatible providers (openai, volcengine, deepseek, dashscope, mistral, gemini)
   const providerDef = BUILT_IN_PROVIDERS[provider];
   if (!providerDef) {
     throw new Error(
@@ -185,7 +221,59 @@ export async function queryLLM(
   }
 
   const apiBase = options.baseUrl || providerDef.baseUrl;
-  const response = await providerFetch(`${apiBase}/chat/completions`, {
+  if (provider === "anthropic") {
+    const response = await providerFetch(
+      `${normalizeProviderBaseUrl(apiBase)}/messages`,
+      {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          "x-api-key": apiKey,
+          "anthropic-version": "2023-06-01",
+        },
+        body: JSON.stringify({
+          model,
+          max_tokens: 4096,
+          messages: providerMessages,
+        }),
+      },
+    );
+    if (!response.ok) {
+      throw new Error(
+        `anthropic error: ${response.status} ${response.statusText}`,
+      );
+    }
+    const data = await response.json();
+    return (
+      data.content
+        ?.filter((part) => part.type === "text")
+        .map((part) => part.text)
+        .join("") || ""
+    );
+  }
+
+  if (provider === "gemini") {
+    const response = await providerFetch(
+      geminiGenerateContentUrl(apiBase, model),
+      {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          "x-goog-api-key": apiKey,
+        },
+        body: JSON.stringify(geminiRequestBody(providerMessages)),
+      },
+    );
+    if (!response.ok) {
+      throw new Error(`gemini error: ${response.status} ${response.statusText}`);
+    }
+    const data = await response.json();
+    return geminiResponseText(data);
+  }
+
+  const response = await providerFetch(
+    `${openAICompatibleBaseUrl(provider, apiBase)}/chat/completions`,
+    {
     method: "POST",
     headers: {
       "Content-Type": "application/json",
@@ -195,7 +283,8 @@ export async function queryLLM(
       model: model || providerDef.models[0],
       messages: providerMessages,
     }),
-  });
+    },
+  );
 
   if (!response.ok) {
     throw new Error(
@@ -252,26 +341,12 @@ export function registerAskCommand(program, dependencies = {}) {
     .option("--json", "Output as JSON")
     .action(async (question, options) => {
       const config = loadConfig();
-      const provider = options.provider || config.llm?.provider || "ollama";
-      const resolvedOptions = {
-        model: options.model || config.llm?.model || "qwen2:7b",
-        provider,
-        apiKey: options.apiKey || config.llm?.apiKey,
-      };
+      const resolvedOptions = resolveAskOptions(options, config);
+      const { provider } = resolvedOptions;
       // baseUrl resolution only applies the Android-local override path for
       // provider=ollama (LocalLlmServer is Ollama-compat). For other providers
       // we use the existing precedence: --base-url > config.llm.baseUrl >
       // provider default.
-      if (provider === "ollama") {
-        resolvedOptions.baseUrl = resolveOllamaBaseUrl({
-          options,
-          env: process.env,
-          config,
-        });
-      } else {
-        resolvedOptions.baseUrl = options.baseUrl || config.llm?.baseUrl;
-      }
-
       // Expand @path references (e.g. `cc ask "summarize @notes.md"`) before
       // sending to the model. The original question is preserved for --json
       // output; only the LLM sees the injected file contents.

package/src/commands/complete.js

@@ -17,6 +17,7 @@ import { randomUUID } from "node:crypto";
 import { types as utilTypes } from "node:util";
 import { captureAgentEvolutionRuntimeComposition } from "../lib/evolution/agent-evolution-runtime-composition-brand.js";
 import { resolveOllamaBaseUrl, queryLLM } from "./ask.js";
+import { resolveLlmConnection } from "../lib/llm-connection-resolution.js";
 
 /** Hard cap so a runaway model can't flood the editor with a whole file. */
 export const MAX_COMPLETION_CHARS = 2000;
@@ -135,19 +136,22 @@ export function registerCompleteCommand(program, dependencies = {}) {
         return;
       }
 
-      const provider = options.provider || config.llm?.provider || "ollama";
+      const connection = resolveLlmConnection(options, [config.llm || {}], {
+        defaultModel:
+          (options.provider || config.llm?.provider || "ollama") === "ollama"
+            ? "qwen2.5-coder"
+            : null,
+      });
+      const provider = connection.provider;
       const resolvedOptions = {
         provider,
-        model:
-          options.model ||
-          config.llm?.model ||
-          (provider === "ollama" ? "qwen2.5-coder" : undefined),
-        apiKey: options.apiKey || config.llm?.apiKey,
+        model: connection.model || (provider === "ollama" ? "qwen2.5-coder" : undefined),
+        apiKey: connection.apiKey,
+        baseUrl: connection.baseUrl,
       };
-      resolvedOptions.baseUrl =
-        provider === "ollama"
-          ? resolveOllamaBaseUrl({ options, config })
-          : options.baseUrl || config.llm?.baseUrl;
+      if (provider === "ollama" && config.llm?.provider === "ollama") {
+        resolvedOptions.baseUrl = resolveOllamaBaseUrl({ options, config });
+      }
 
       try {
         const prompt = buildFimPrompt(prefix, suffix, language);

package/src/commands/llm.js

@@ -43,6 +43,7 @@ import {
   autoFailStuckRequestsV2,
   getLlmProvidersStatsV2,
 } from "../lib/llm-providers.js";
+import { normalizeOllamaBaseUrl } from "../lib/ollama-base-url.js";
 import {
   readLlmConnectionInput,
   saveLlmConnection,
@@ -52,6 +53,7 @@ import {
   llmProbeReceipt,
 } from "../lib/llm-connection-probe.js";
 import { resolveModelCapabilityProfile } from "../lib/model-capabilities.js";
+import { resolveLlmConnection } from "../lib/llm-connection-resolution.js";
 
 /**
  * Resolve the effective `cc llm test` target from CLI flags + persisted config.
@@ -75,46 +77,27 @@ export function resolveLlmTestTarget(
 ) {
   const llm = config.llm || {};
   const provider = options.provider || llm.provider || "ollama";
-  const sameProvider = provider === (llm.provider || "ollama");
   const isOllama = provider === "ollama";
   const builtIn = builtIns[provider];
-
-  const model =
-    options.model ||
-    (sameProvider ? llm.model : undefined) ||
-    (isOllama ? "qwen2:7b" : builtIn?.models?.[0] || "gpt-4o-mini");
-
-  let baseUrl;
-  if (isOllama) {
-    // Only inherit config.baseUrl when the CONFIGURED provider is also ollama —
-    // otherwise `--provider ollama` would probe another provider's endpoint.
-    const configOllamaBase =
-      llm.provider === "ollama" ? llm.baseUrl : undefined;
-    baseUrl = options.baseUrl || configOllamaBase || "http://localhost:11434";
-  } else {
-    baseUrl =
-      options.baseUrl ||
-      (sameProvider ? llm.baseUrl : undefined) ||
-      builtIn?.baseUrl ||
-      "https://api.openai.com/v1";
-  }
-
-  const apiKey = isOllama
-    ? undefined
-    : options.apiKey ||
-      (sameProvider ? llm.apiKey : undefined) ||
-      (builtIn?.apiKeyEnv ? env[builtIn.apiKeyEnv] : undefined);
-
+  const connection = resolveLlmConnection(options, [llm], {
+    env,
+    builtIns,
+    defaultProvider: provider,
+    defaultModel: isOllama
+      ? "qwen2:7b"
+      : builtIn?.models?.[0] || "gpt-4o-mini",
+  });
   return {
-    provider,
-    model,
-    baseUrl,
-    apiKey,
+    provider: connection.provider,
+    model: connection.model,
+    baseUrl:
+      connection.baseUrl ||
+      (isOllama ? "http://localhost:11434" : "https://api.openai.com/v1"),
+    apiKey: isOllama ? undefined : connection.apiKey || undefined,
     isOllama,
     label: builtIn?.displayName || (isOllama ? "Ollama" : provider),
   };
 }
-
 /** Resolve a static profile without reading credentials or provider key envs. */
 export function resolveLlmCapabilities(
   options = {},
@@ -253,7 +236,9 @@ export function registerLlmCommand(program) {
       const spinner = ora("Fetching models...").start();
       try {
         if (options.provider === "ollama") {
-          const response = await fetch(`${options.baseUrl}/api/tags`);
+          const response = await fetch(
+            `${normalizeOllamaBaseUrl(options.baseUrl)}/api/tags`,
+          );
           if (!response.ok) {
             throw new Error(`Ollama not reachable: ${response.status}`);
           }
@@ -300,13 +285,10 @@ export function registerLlmCommand(program) {
 
   // llm test - test connectivity
   //
-  // Defaults to the CONFIGURED provider/model/baseUrl/apiKey (config.llm.*),
-  // NOT a hard-coded ollama. Previously this always probed ollama unless you
-  // passed --provider, so a user on volcengine/openai/etc. (and the JetBrains
-  // "Configure LLM" wizard, which calls `cc llm test`) got a spurious
-  // "fetch failed" against a local ollama they never ran. Any non-ollama
-  // built-in provider is OpenAI-compatible, so they share one /chat/completions
-  // probe instead of only special-casing openai.
+  // Defaults to the configured provider/model/baseUrl/apiKey (config.llm.*),
+  // not a hard-coded Ollama endpoint. The probe routes Ollama, Anthropic, and
+  // Gemini through their native protocols and uses chat/completions for the
+  // remaining OpenAI-compatible providers.
   llm
     .command("test")
     .description(

package/src/commands/session.js

@@ -760,9 +760,9 @@ export function registerSessionCommand(program, dependencies = {}) {
         const ingress = composition.evolutionIngress;
         await ingress.start();
         const lifecycle = await startChatRepl({
-          model: options.model || sess.model || "qwen2:7b",
+          model: options.model || sess.model,
           provider: options.provider || sess.provider || "ollama",
-          baseUrl: options.baseUrl || "http://localhost:11434",
+          baseUrl: options.baseUrl,
           resumeMessages: sess.messages,
           sessionId: sess.id,
           evolutionIngress: ingress,

package/src/commands/stream.js

@@ -13,6 +13,7 @@
 import { logger } from "../lib/logger.js";
 import { loadConfig } from "../lib/config-manager.js";
 import { buildProviderSource } from "../lib/provider-stream.js";
+import { resolveLlmConnection } from "../lib/llm-connection-resolution.js";
 import {
   readEvolutionCompositionFactory,
   prepareGovernedModelTurn,
@@ -29,16 +30,19 @@ export function registerStreamCommand(program, dependencies = {}) {
     )
     .argument("<prompt>", "The prompt to stream")
     .option("--model <model>", "Model name")
-    .option("--provider <provider>", "LLM provider", "ollama")
+    .option("--provider <provider>", "LLM provider (default: configured)")
     .option("--base-url <url>", "API base URL")
     .option("--api-key <key>", "API key")
     .option("--text", "Emit concatenated final text instead of NDJSON events")
     .action(async (prompt, options) => {
       const config = loadConfig();
-      const provider = options.provider || config.llm?.provider || "ollama";
-      const baseUrl = options.baseUrl || config.llm?.baseUrl;
-      const apiKey = options.apiKey || config.llm?.apiKey;
-      const model = options.model || config.llm?.model;
+      const connection = resolveLlmConnection(options, [config.llm || {}], {
+        defaultModel:
+          (options.provider || config.llm?.provider || "ollama") === "ollama"
+            ? "qwen2:7b"
+            : null,
+      });
+      const { provider, baseUrl, apiKey, model } = connection;
 
       try {
         const { createStreamRouter } =

package/src/commands/ui.js

@@ -1,6 +1,8 @@
 import { logger } from "../lib/logger.js";
 import { createAgentRuntimeFactory } from "../runtime/runtime-factory.js";
 import { readEvolutionCompositionFactory } from "../lib/evolution/governed-model-turn.js";
+import { loadEvolutionDeploymentCommandDependencies } from "../lib/evolution/evolution-deployment-loader.js";
+import { getEvolutionDeploymentStatus } from "../lib/evolution/evolution-deployment-config.js";
 
 export function registerUiCommand(program, dependencies = {}) {
   const evolutionCompositionFactory =
@@ -27,6 +29,29 @@ export function registerUiCommand(program, dependencies = {}) {
     )
     .action(async (opts) => {
       try {
+        let activeEvolutionCompositionFactory = evolutionCompositionFactory;
+        if (activeEvolutionCompositionFactory === null) {
+          let deploymentDependencies = null;
+          let localTrustSetupError = null;
+          try {
+            deploymentDependencies =
+              await loadEvolutionDeploymentCommandDependencies("ui", {
+                initializeDefaultLocalUi: true,
+              });
+          } catch (error) {
+            localTrustSetupError = error;
+          }
+          activeEvolutionCompositionFactory = readEvolutionCompositionFactory(
+            deploymentDependencies || {},
+          );
+          if (activeEvolutionCompositionFactory === null) {
+            const status = await getEvolutionDeploymentStatus();
+            const readiness = status.readiness.ui;
+            logger.warn(
+              `Quick Ask model egress is unavailable (${readiness.code}): ${localTrustSetupError?.message || readiness.detail}. ${readiness.remediation || "Review cc evolution deployment status."}`,
+            );
+          }
+        }
         // Allow env vars to override CLI defaults — host/token/port/uiMode.
         // Use case: Android local terminal app pre-injects CC_UI_HOST=0.0.0.0
         // + a per-install CC_UI_TOKEN so `cc ui` is reachable from other
@@ -47,9 +72,12 @@ export function registerUiCommand(program, dependencies = {}) {
         );
         const runtime = createAgentRuntimeFactory({
           deps:
-            evolutionCompositionFactory === null
+            activeEvolutionCompositionFactory === null
               ? {}
-              : { evolutionCompositionFactory },
+              : {
+                  evolutionCompositionFactory:
+                    activeEvolutionCompositionFactory,
+                },
         }).createUiRuntime({
           port,
           wsPort: parseInt(opts.wsPort, 10),

package/src/data/changelog.json

@@ -1,6 +1,58 @@
 {
   "source": "CHANGELOG.md",
   "releases": [
+    {
+      "productVersion": null,
+      "cliVersion": "0.166.112",
+      "cliVersions": ["0.166.112"],
+      "date": "2026-10-21",
+      "title": "",
+      "sections": [
+        {
+          "heading": "Fixed - cc CLI 0.166.112: retain generated Evolution test signing keys (candidate, 2026-10-21)",
+          "body": "- Preserve the supplied private signing-key path when saving an explicit Evolution test deployment, so generated test deployments remain usable after configuration.\n- Align Windows ACL fixtures with Session Core's extended-path PowerShell protocol while keeping CLI-facing paths unchanged.\n- Pair VS Code 0.37.155 and JetBrains 0.4.172 with CLI 0.166.112. Publish Session Core 0.3.21 first, then CLI through the exact-SHA GitHub Actions OIDC workflow; publish IDE extensions only after npm readback succeeds."
+        }
+      ]
+    },
+    {
+      "productVersion": null,
+      "cliVersion": "0.166.111",
+      "cliVersions": ["0.166.111"],
+      "date": "2026-10-21",
+      "title": "",
+      "sections": [
+        {
+          "heading": "Fixed - cc CLI 0.166.111: route every configured provider correctly (candidate, 2026-10-21)",
+          "body": "- Route Gemini Quick Ask, `cc ask`, interactive chat and `cc stream` through Google's native `generateContent` API; keep agent and Hub compatibility clients on Google's `/v1beta/openai` endpoint.\n- Send Anthropic through its native Messages API in `cc ask` and `cc stream`; keep OpenAI-compatible providers and configured proxies on `/chat/completions`.\n- Remove trailing slashes before provider endpoint paths, send Gemini credentials in headers, and avoid reusing a different provider's saved model, URL or API key after a provider override.\n- Publish Session Core 0.3.21 before this CLI release for the Windows long-path ACL fix used by signed local Quick Ask storage.\n- Pair IDE candidates VS Code 0.37.155 and JetBrains 0.4.172 with CLI 0.166.111 after the public CLI package is available."
+        }
+      ]
+    },
+    {
+      "productVersion": null,
+      "cliVersion": "0.166.110",
+      "cliVersions": ["0.166.110"],
+      "date": "2026-10-21",
+      "title": "",
+      "sections": [
+        {
+          "heading": "Fixed - cc CLI 0.166.110: normalize Ollama native API URLs (candidate, 2026-10-21)",
+          "body": "- Strip an optional trailing `/v1` from Ollama base URLs before native `/api/*` requests across agent chat, streaming chat, model probes, and provider checks, preventing requests such as `/v1/api/chat` from returning 404.\n- Save Ollama connection settings with the native API base URL while preserving `/v1` on OpenAI-compatible providers.\n- Initialize a user-scoped signed Evolution host on the first `cc ui` launch when no deployment exists, removing manual signature setup from Quick Ask while keeping the local authority limited to UI model turns and bound to each run.\n- Add a verified Windows local NTFS directory flush for local UI evidence before witness CAS; unsupported profile storage fails during initialization rather than after starting Quick Ask.\n- Child npm package versions remain unchanged. IDE extensions are not part of this CLI release."
+        }
+      ]
+    },
+    {
+      "productVersion": null,
+      "cliVersion": "0.166.109",
+      "cliVersions": ["0.166.109"],
+      "date": "2026-10-21",
+      "title": "",
+      "sections": [
+        {
+          "heading": "Fixed - cc CLI 0.166.109: use saved LLM settings for Web Panel Quick Ask (candidate, 2026-10-21)",
+          "body": "- Resolve the selected provider, endpoint, model, and API key from current saved CLI settings for Web Panel Quick Ask requests.\n- Avoid pairing credentials with a different saved endpoint during concurrent config updates, and keep provider/environment fallback behavior explicit.\n- Normalize Ollama endpoints that include the OpenAI-compatible `/v1` suffix before sending Ollama requests.\n- Child npm package versions remain unchanged. IDE extensions are not part of this CLI release."
+        }
+      ]
+    },
     {
       "productVersion": null,
       "cliVersion": "0.166.108",

package/src/gateways/ws/llm-chat-protocol.js

@@ -26,7 +26,8 @@
  * Provider resolution (first non-null wins):
  *   1. message.options — explicit provider/model/apiKey from the client
  *   2. server.sessionManager.getSession(sessionId) — set via `cc auth llm`
- *   3. process.env[BUILT_IN_PROVIDERS[provider].apiKeyEnv]
+ *   3. Current saved `config.llm`
+ *   4. process.env[BUILT_IN_PROVIDERS[provider].apiKeyEnv]
  *
  * Without any source of credentials we send a clean ok:false frame instantly
  * — the user sees an actionable error rather than a 60s hang.
@@ -36,10 +37,13 @@ import {
   streamOllama,
   streamOpenAI,
   streamAnthropic,
+  streamGemini,
 } from "../../lib/chat-core.js";
 import { BUILT_IN_PROVIDERS } from "../../lib/llm-providers.js";
 import { resolveLlmCreds } from "./llm-creds.js";
 import { prepareGovernedModelTurn } from "../../lib/evolution/governed-model-turn.js";
+import { normalizeOllamaBaseUrl } from "../../lib/ollama-base-url.js";
+import { normalizeProviderBaseUrl } from "../../lib/provider-protocol.js";
 
 function validateMessages(messages) {
   if (!Array.isArray(messages) || messages.length === 0)
@@ -126,7 +130,10 @@ export async function handleLlmChat(server, id, ws, message) {
       requireCompletion: turn.governed,
       signal: controller.signal,
     };
-    const baseUrl = creds.baseUrl || "http://localhost:11434";
+    const baseUrl =
+      creds.provider === "ollama"
+        ? normalizeOllamaBaseUrl(creds.baseUrl)
+        : normalizeProviderBaseUrl(creds.baseUrl);
     if (creds.provider === "ollama") {
       await streamOllama(
         turn.messages,
@@ -138,8 +145,7 @@ export async function handleLlmChat(server, id, ws, message) {
         streamOptions,
       );
     } else if (creds.provider === "anthropic") {
-      const def = BUILT_IN_PROVIDERS.anthropic;
-      const url = baseUrl !== "http://localhost:11434" ? baseUrl : def.baseUrl;
+      const url = baseUrl;
       await streamAnthropic(
         turn.messages,
         creds.model,
@@ -150,12 +156,24 @@ export async function handleLlmChat(server, id, ws, message) {
         undefined,
         streamOptions,
       );
+    } else if (creds.provider === "gemini") {
+      const url = baseUrl;
+      await streamGemini(
+        turn.messages,
+        creds.model,
+        url,
+        creds.apiKey,
+        onToken,
+        onUsage,
+        undefined,
+        streamOptions,
+      );
     } else {
       // OpenAI-compatible — covers openai / volcengine / deepseek / dashscope /
       // kimi / minimax / mistral / custom. baseUrl from creds; default-from
       // BUILT_IN_PROVIDERS already applied above.
       const def = BUILT_IN_PROVIDERS[creds.provider];
-      const url = baseUrl !== "http://localhost:11434" ? baseUrl : def?.baseUrl;
+      const url = baseUrl || def?.baseUrl;
       await streamOpenAI(
         turn.messages,
         creds.model,