copilot-relay 0.3.3 → 0.3.5

package/README.md

@@ -49,13 +49,14 @@ Routing:
 | Requested model | Upstream model |
 | --- | --- |
 | contains `opus` | `claude-opus-5` |
-| `gpt-5.6-sol[1m]`, plain `gpt-5.6-sol`, or another non-Opus alias | `gpt-5.6-sol` |
+| `gpt-6-astra[1m]`, plain `gpt-6-astra`, or another non-Opus alias | `gpt-6-astra` |
 
-The relay advertises `gpt-5.6-sol[1m]` to Claude Code. The `[1m]` selector
-changes Claude Code's client-side context budgeting only; every Copilot request
-still uses `gpt-5.6-sol`, and the relay cannot enlarge GitHub Copilot's upstream
-capacity. An explicit CLI or API model override that bypasses the managed Claude
-settings is outside this guarantee.
+The fresh-install default is `gpt-6-astra`, exposed to Claude Code as
+`gpt-6-astra[1m]`. The selector changes Claude Code's client-side context budgeting
+only; Copilot receives the canonical `gpt-6-astra` ID. It does not enlarge
+upstream capacity. Existing `config.yaml` model choices are never migrated.
+If your account cannot use Astra, startup fails explicitly; choose a model your
+account supports rather than relying on an automatic fallback.
 
 Claude Code can still show its built-in Haiku and Sonnet picker entries. The
 managed `model` field sets the startup default but does not restrict
@@ -73,8 +74,8 @@ npx copilot-relay@latest stop
 With `claudeSetup: true`, `start` manages `ANTHROPIC_BASE_URL`, a dummy
 `ANTHROPIC_AUTH_TOKEN` when auth is absent, and the configured GPT default via
 the top-level `model` field in `~/.claude/settings.json`. Exact
-`gpt-5.6-sol` model overrides are normalized to the Claude-facing
-`gpt-5.6-sol[1m]` identity; unrelated model choices are preserved.
+`gpt-6-astra` and `gpt-5.6-sol` overrides are normalized to Claude-facing
+`[1m]` identities; unrelated model choices are preserved.
 
 ## Config
 
@@ -90,12 +91,16 @@ logRetentionDays: 3
 thinkEffort: max
 upstreamTimeoutSeconds: 180
 webSearchBackend:
+gptModel: gpt-6-astra
+opusModel: claude-opus-5
 ```
 
 copilot-relay writes the resolved config back to this file, so every key is
 present after the first start. Shipped defaults therefore apply to fresh
 installs only — once a value is in your `config.yaml` it is never rewritten by
-an upgrade. To pick up a changed default, edit the key yourself.
+an upgrade. To pick up a changed default, edit the key yourself. See
+[Configuration](wiki/EN-Configuration.md) / [配置说明](wiki/ZH-Configuration.md)
+for live model discovery, compatible thinking efforts, and copy-paste updates.
 
 `logLevel` controls verbosity:
 
@@ -137,7 +142,7 @@ copilot-relay 0.2.5
   version    0.2.5
   listening  http://127.0.0.1:4142
   health     ok (9ms)
-  models     gpt-5.6-sol[1m], claude-opus-5
+  models     gpt-6-astra[1m], claude-opus-5
   upstream   not checked (use --deep)
   log        ~/.copilot-relay/logs/copilot-relay.2026-07-25.log
   config     ~/.copilot-relay/config.yaml
@@ -150,7 +155,7 @@ copilot-relay 0.2.5
     thinkEffort             max
     upstreamTimeoutSeconds  180
     webSearchBackend        (unset — uses gptModel)
-    gptModel                gpt-5.6-sol
+    gptModel                gpt-6-astra
     opusModel               claude-opus-5
     host, port and claudeSetup take effect on restart; the rest hot-reload.
 ```

package/config.default.yaml

@@ -52,7 +52,7 @@ webSearchBackend:
 # Model routing: requests containing "opus" use opusModel; all others use gptModel.
 #
 # Upstream Copilot model used for non-Opus requests.
-gptModel: gpt-5.6-sol
+gptModel: gpt-6-astra
 
 # Upstream Copilot model used for requests containing "opus".
 opusModel: claude-opus-5

package/dist/main.js

@@ -543,15 +543,19 @@ const runtimeState = {};
 //#region src/lib/models.ts
 const defaultReasoningEffort = "max";
 const defaultModelRouting = {
-	gptModel: "gpt-5.6-sol",
+	gptModel: "gpt-6-astra",
 	opusModel: "claude-opus-5"
 };
 const isReasoningEffort = (value) => value === "none" || value === "low" || value === "medium" || value === "high" || value === "xhigh" || value === "max";
-const exactGpt56SolPattern = /^gpt-5\.6-sol(?:\[1m\])*$/i;
-const claudeContextModelId = "gpt-5.6-sol[1m]";
-const copilotModelId = "gpt-5.6-sol";
-const normalizeClaudeModelId = (model) => exactGpt56SolPattern.test(model) ? claudeContextModelId : model;
-const normalizeCopilotModelId = (model) => exactGpt56SolPattern.test(model) ? copilotModelId : model;
+const oneMillionContextModelPattern = /^(gpt-5\.6-sol|gpt-6-astra)(?:\[1m\])*$/i;
+const normalizeOneMillionContextModel = (model) => {
+	return oneMillionContextModelPattern.exec(model)?.[1]?.toLowerCase();
+};
+const normalizeClaudeModelId = (model) => {
+	const normalized = normalizeOneMillionContextModel(model);
+	return normalized ? `${normalized}[1m]` : model;
+};
+const normalizeCopilotModelId = (model) => normalizeOneMillionContextModel(model) ?? model;
 const getConfiguredModelRouting = () => runtimeState.modelRouting ?? defaultModelRouting;
 const getModelRouting = () => {
 	const routing = getConfiguredModelRouting();
@@ -591,7 +595,7 @@ const logLevels = [
 const defaultConfig = {
 	claudeSetup: true,
 	copilotBaseUrl: "https://api.githubcopilot.com",
-	gptModel: "gpt-5.6-sol",
+	gptModel: "gpt-6-astra",
 	host: "127.0.0.1",
 	logLevel: "info",
 	logRetentionDays: 3,
@@ -1197,7 +1201,7 @@ const hasPrimaryModelOverride = (parsed, env) => typeof parsed.model === "string
 /**
 * Update `~/.claude/settings.json` so its env block points Claude Code at the
 * running proxy. Preserves unrelated keys and model choices while normalizing
-* managed gpt-5.6-sol overrides. Sets a dummy auth token only if none is present.
+* managed one-million-context GPT overrides. Sets a dummy auth token only if absent.
 */
 async function applyClaudeConfig(input) {
 	const { configPath: configPath$1, baseUrl, gptModel } = input;
@@ -1387,7 +1391,7 @@ const fetchCopilot = async (provider, path$1, init, options = {}) => {
 
 //#endregion
 //#region src/copilot/responses.ts
-const responsesOnlyModelPattern = /^(?:gpt-5\.5|gpt-5\.6)(?:-|$)/i;
+const responsesOnlyModelPattern = /^(?:gpt-5\.5|gpt-5\.6|gpt-6-astra)(?:-|$)/i;
 function shouldUseResponsesApiForModel(model) {
 	return responsesOnlyModelPattern.test(model);
 }
@@ -3646,7 +3650,7 @@ const renderStatus = (status$1) => {
 	const health = status$1.health;
 	lines.push(health?.ok ? `  health     ok (${health.ms}ms)` : `  health     FAILED${health?.detail ? ` — ${health.detail}` : ""}`);
 	lines.push(status$1.models?.length ? `  models     ${status$1.models.join(", ")}` : "  models     unavailable");
-	if (status$1.deep) lines.push(status$1.deep.ok ? `  upstream   ok (${status$1.deep.ms}ms) — end-to-end Copilot round trip` : `  upstream   FAILED${status$1.deep.detail ? ` — ${status$1.deep.detail}` : ""}`);
+	if (status$1.deep) lines.push(status$1.deep.ok ? `  upstream   ok (${status$1.deep.ms}ms) — ${status$1.deep.detail ?? "end-to-end Copilot round trip"}` : `  upstream   FAILED${status$1.deep.detail ? ` — ${status$1.deep.detail}` : ""}`);
 	else lines.push("  upstream   not checked (use --deep)");
 	lines.push(`  log        ${status$1.logPath}`, `  config     ${status$1.configPath}`, ...renderConfig(status$1.config, true));
 	if (status$1.health?.ok && status$1.deep && !status$1.deep.ok) lines.push("", "  The relay is listening but cannot reach Copilot.", "  Try: copilot-relay auth");
@@ -3736,11 +3740,19 @@ const checkDeep = async (baseUrl, model) => {
 				ok: false
 			};
 		}
-		const content = result.body?.content;
-		return Array.isArray(content) && content.length > 0 ? {
-			ms: result.ms,
-			ok: true
-		} : {
+		const body = result.body;
+		if (Array.isArray(body?.content)) {
+			if (body.content.length > 0) return {
+				ms: result.ms,
+				ok: true
+			};
+			if (body.type === "message" && body.role === "assistant" && body.stop_reason === "max_tokens" && typeof body.usage?.output_tokens === "number" && Number.isFinite(body.usage.output_tokens) && body.usage.output_tokens > 0) return {
+				detail: "output budget exhausted before visible text",
+				ms: result.ms,
+				ok: true
+			};
+		}
+		return {
 			detail: "empty response",
 			ms: result.ms,
 			ok: false