@protolabsai/proto 0.29.0 → 0.31.0

package/README.md

@@ -27,7 +27,7 @@ At-a-glance overview vs. upstream Qwen Code. For the full architectural breakdow
 | Ignore files          | `.qwenignore`             | `.protoignore` + inherits `.claudeignore` patterns                                                                                                       |
 | ACP / Zed integration | Stock                     | Cron-in-Session, concurrent Agent calls, SSE/HTTP MCP, internal-part filtering                                                                           |
 | Extra built-in tools  | Standard set              | + browser automation, repo-map (PageRank), task tools, mailbox, LSP, voice/STT                                                                           |
-| Observability         | Console                   | Langfuse OTLP traces with harness-intervention spans (SFT-ready)                                                                                         |
+| Observability         | Console                   | OTLP/HTTP to LGTM stack + Langfuse, opt-in, with `gen_ai.response.thinking` and harness-intervention spans (SFT-ready)                                   |
 | Release pipeline      | Manual                    | Conventional-commit auto-release (`feat:` → minor, `fix:` → patch)                                                                                       |
 | VS Code companion     | Included                  | Removed (focus on TUI + ACP/Zed)                                                                                                                         |
 
@@ -206,53 +206,56 @@ Both no-op outside a TTY, in screen-reader mode, or under tmux/SSH.
 
 ## Observability
 
-proto supports [Langfuse](https://langfuse.com) tracing out of the box. Set three environment variables and every session is fully traced — LLM calls (all providers), tool executions, subagent lifecycles, and turn hierarchy.
+proto ships OpenTelemetry-native, with both a Tempo/LGTM-style ops backend and Langfuse for prompt-grade trace UI. Both are **opt-in** — nothing is sent anywhere until `telemetry.enabled` is `true`.
 
 ### Setup
 
-Add to the `env` block in `~/.proto/settings.json`:
+Add to `~/.proto/settings.json`:
 
 ```json
 {
+  "telemetry": { "enabled": true },
   "env": {
+    "OTEL_INGRESS_TOKEN": "<bearer token from your Infisical or vault>",
     "LANGFUSE_PUBLIC_KEY": "pk-lf-...",
     "LANGFUSE_SECRET_KEY": "sk-lf-...",
-    "LANGFUSE_BASE_URL": "https://cloud.langfuse.com"
+    "LANGFUSE_BASE_URL": "https://your-langfuse-instance.example.com"
   }
 }
 ```
 
-`LANGFUSE_BASE_URL` is optional and defaults to `https://cloud.langfuse.com`. For a self-hosted instance, set it to your deployment URL.
+With `telemetry.enabled = true`:
 
-> **Why `settings.json` and not `.env`?** proto walks up from your CWD loading `.env` files, so a project-level `.env` with Langfuse keys would bleed into proto's tracing and mix your traces into the wrong dataset. The `env` block in `settings.json` is proto-namespaced and completely isolated from your projects.
+- **OTLP traces** ship to `https://otel.proto-labs.ai` over HTTP, bearer-auth via `OTEL_INGRESS_TOKEN`. Override `telemetry.otlpEndpoint` / `telemetry.otlpProtocol` to point at a local OTel collector or a different vendor.
+- **Langfuse traces** ship to `LANGFUSE_BASE_URL` (defaults to `https://cloud.langfuse.com`) when both Langfuse keys are present.
+
+Without `telemetry.enabled = true`, neither exporter activates regardless of env vars.
+
+> **Why `settings.json` and not `.env`?** proto walks up from your CWD loading `.env` files, so a project-level `.env` with telemetry keys would bleed into proto's tracing and mix your traces into the wrong dataset. The `env` block in `settings.json` is proto-namespaced and completely isolated from your projects.
 
 ### What gets traced
 
-| Span                  | Attributes                                                                                           |
-| --------------------- | ---------------------------------------------------------------------------------------------------- |
-| `turn`                | `session.id`, `turn.id` — root span per user prompt                                                  |
-| `gen_ai chat {model}` | `gen_ai.usage.input_tokens`, `gen_ai.usage.output_tokens`, `gen_ai.request.model` — one per LLM call |
-| `tool/{name}`         | `tool.name`, `tool.type`, `tool.duration_ms` — one per tool execution                                |
-| `agent/{name}`        | `agent.name`, `agent.status`, `agent.duration_ms` — one per subagent                                 |
+| Span                  | Attributes                                                                                                                          |
+| --------------------- | ----------------------------------------------------------------------------------------------------------------------------------- |
+| `turn`                | `session.id`, `turn.id` — root span per user prompt                                                                                 |
+| `gen_ai chat {model}` | `gen_ai.usage.{input,output,thinking}_tokens`, `gen_ai.request.model`, `gen_ai.response.thinking` (when present) — one per LLM call |
+| `tool/{name}`         | `tool.name`, `tool.type`, `tool.duration_ms` — one per tool execution                                                               |
+| `agent/{name}`        | `agent.name`, `agent.status`, `agent.duration_ms` — one per subagent                                                                |
 
 All three provider backends are covered: OpenAI-compatible, Anthropic, and Gemini.
 
 ### Prompt content logging
 
-Full prompt messages and response text are included in traces by default. To disable:
+Full prompt messages, response text, and reasoning text are included in traces by default. To disable:
 
 ```json
 // ~/.proto/settings.json
 {
-  "telemetry": { "logPrompts": false }
+  "telemetry": { "enabled": true, "logPrompts": false }
 }
 ```
 
-> **Privacy note:** `logPrompts` is enabled by default. When enabled, full prompt and response content is sent to your Langfuse instance. Set to `false` if you want traces without message content.
-
-### Langfuse activates independently
-
-Langfuse tracing activates from env vars alone — it does not require `telemetry.enabled: true` in settings. The general telemetry pipeline (OTLP/GCP) and Langfuse are independent.
+> **Privacy note:** Telemetry is off by default. When you opt in, `logPrompts` defaults to `true` — full prompt, response, and reasoning content are attached to spans (truncated at 10K chars each). Set `logPrompts: false` if you want token counts and timings without message content.
 
 ## Task Management
 

package/cli.js

@@ -93610,7 +93610,7 @@ var require_metadata3 = __commonJS({
       }
     }
     __name(validate3, "validate");
-    var Metadata = class _Metadata {
+    var Metadata2 = class _Metadata {
       static {
         __name(this, "Metadata");
       }
@@ -93782,7 +93782,7 @@ var require_metadata3 = __commonJS({
         return result;
       }
     };
-    exports2.Metadata = Metadata;
+    exports2.Metadata = Metadata2;
     var bufToString = /* @__PURE__ */ __name((val) => {
       return Buffer.isBuffer(val) ? val.toString("base64") : val;
     }, "bufToString");
@@ -116794,10 +116794,10 @@ var require_grpc_exporter_transport = __commonJS({
     exports2.createSslCredentials = createSslCredentials;
     function createEmptyMetadata() {
       const {
-        Metadata
+        Metadata: Metadata2
         // eslint-disable-next-line @typescript-eslint/no-require-imports
       } = require_src11();
-      return new Metadata();
+      return new Metadata2();
     }
     __name(createEmptyMetadata, "createEmptyMetadata");
     exports2.createEmptyMetadata = createEmptyMetadata;
@@ -141044,11 +141044,14 @@ function parseOtlpEndpoint(otlpEndpointSetting, protocol) {
   }
 }
 function initializeTelemetry(config2) {
-  const langfuse = buildLangfuseExporters();
-  if (telemetryInitialized || !config2.getTelemetryEnabled() && !langfuse) {
+  const debugLogger164 = createDebugLogger("OTEL");
+  if (telemetryInitialized || !config2.getTelemetryEnabled()) {
+    if (!telemetryInitialized && (process.env["LANGFUSE_PUBLIC_KEY"] || process.env["LANGFUSE_SECRET_KEY"])) {
+      debugLogger164.debug("Langfuse env vars detected but telemetry.enabled is false \u2014 skipping. Set telemetry.enabled = true in settings to opt in.");
+    }
     return;
   }
-  const debugLogger164 = createDebugLogger("OTEL");
+  const langfuse = buildLangfuseExporters();
   const resource = (0, import_resources.resourceFromAttributes)({
     [SemanticResourceAttributes.SERVICE_NAME]: SERVICE_NAME,
     [SemanticResourceAttributes.SERVICE_VERSION]: process.version,
@@ -141066,32 +141069,49 @@ function initializeTelemetry(config2) {
   let logExporter;
   let metricReader;
   if (useOtlp) {
+    const otlpAuthToken = process.env["OTEL_INGRESS_TOKEN"];
+    const otlpHeaders = otlpAuthToken ? { Authorization: `Bearer ${otlpAuthToken}` } : void 0;
+    if (!otlpAuthToken && /otel\.proto-labs\.ai/.test(parsedEndpoint ?? "")) {
+      debugLogger164.debug("OTEL_INGRESS_TOKEN not set; OTLP exports to otel.proto-labs.ai will return 401.");
+    }
     if (otlpProtocol === "http") {
+      const httpAuth = otlpHeaders ? { headers: otlpHeaders } : {};
       spanExporter = new import_exporter_trace_otlp_http.OTLPTraceExporter({
-        url: parsedEndpoint
+        url: parsedEndpoint,
+        ...httpAuth
       });
       logExporter = new import_exporter_logs_otlp_http.OTLPLogExporter({
-        url: parsedEndpoint
+        url: parsedEndpoint,
+        ...httpAuth
       });
       metricReader = new import_sdk_metrics2.PeriodicExportingMetricReader({
         exporter: new import_exporter_metrics_otlp_http.OTLPMetricExporter({
-          url: parsedEndpoint
+          url: parsedEndpoint,
+          ...httpAuth
         }),
         exportIntervalMillis: 1e4
       });
     } else {
+      const grpcAuth = otlpAuthToken ? (() => {
+        const m3 = new import_grpc_js.Metadata();
+        m3.set("authorization", `Bearer ${otlpAuthToken}`);
+        return { metadata: m3 };
+      })() : {};
       spanExporter = new import_exporter_trace_otlp_grpc.OTLPTraceExporter({
         url: parsedEndpoint,
-        compression: CompressionAlgorithm.GZIP
+        compression: CompressionAlgorithm.GZIP,
+        ...grpcAuth
       });
       logExporter = new import_exporter_logs_otlp_grpc.OTLPLogExporter({
         url: parsedEndpoint,
-        compression: CompressionAlgorithm.GZIP
+        compression: CompressionAlgorithm.GZIP,
+        ...grpcAuth
       });
       metricReader = new import_sdk_metrics2.PeriodicExportingMetricReader({
         exporter: new import_exporter_metrics_otlp_grpc.OTLPMetricExporter({
           url: parsedEndpoint,
-          compression: CompressionAlgorithm.GZIP
+          compression: CompressionAlgorithm.GZIP,
+          ...grpcAuth
         }),
         exportIntervalMillis: 1e4
       });
@@ -141158,7 +141178,7 @@ async function shutdownTelemetry() {
     telemetryInitialized = false;
   }
 }
-var import_exporter_trace_otlp_grpc, import_exporter_logs_otlp_grpc, import_exporter_metrics_otlp_grpc, import_exporter_trace_otlp_http, import_exporter_logs_otlp_http, import_exporter_metrics_otlp_http, import_sdk_node, import_resources, import_sdk_trace_node, import_sdk_logs, import_sdk_metrics2, import_instrumentation_http, sdk, telemetryInitialized;
+var import_exporter_trace_otlp_grpc, import_exporter_logs_otlp_grpc, import_exporter_metrics_otlp_grpc, import_exporter_trace_otlp_http, import_exporter_logs_otlp_http, import_exporter_metrics_otlp_http, import_grpc_js, import_sdk_node, import_resources, import_sdk_trace_node, import_sdk_logs, import_sdk_metrics2, import_instrumentation_http, sdk, telemetryInitialized;
 var init_sdk = __esm({
   "packages/core/dist/src/telemetry/sdk.js"() {
     "use strict";
@@ -141171,6 +141191,7 @@ var init_sdk = __esm({
     import_exporter_logs_otlp_http = __toESM(require_src21(), 1);
     import_exporter_metrics_otlp_http = __toESM(require_src18(), 1);
     init_esm3();
+    import_grpc_js = __toESM(require_src11(), 1);
     import_sdk_node = __toESM(require_src34(), 1);
     init_esm2();
     import_resources = __toESM(require_src13(), 1);
@@ -155923,7 +155944,7 @@ var init_pipeline = __esm({
         this.converter = new OpenAIContentConverter(this.contentGeneratorConfig.model, this.contentGeneratorConfig.schemaCompliance, this.contentGeneratorConfig.modalities ?? {});
       }
       async execute(request3, userPromptId) {
-        const effectiveModel = this.contentGeneratorConfig.model;
+        const effectiveModel = this.resolveEffectiveModel(request3);
         this.converter.setModel(effectiveModel);
         this.converter.setModalities(this.contentGeneratorConfig.modalities ?? {});
         return this.executeWithErrorHandling(request3, userPromptId, false, effectiveModel, async (openaiRequest) => {
@@ -155935,7 +155956,7 @@ var init_pipeline = __esm({
         });
       }
       async executeStream(request3, userPromptId) {
-        const effectiveModel = this.contentGeneratorConfig.model;
+        const effectiveModel = this.resolveEffectiveModel(request3);
         this.converter.setModel(effectiveModel);
         this.converter.setModalities(this.contentGeneratorConfig.modalities ?? {});
         return this.executeWithErrorHandling(request3, userPromptId, true, effectiveModel, async (openaiRequest, context2) => {
@@ -156285,6 +156306,22 @@ var init_pipeline = __esm({
         context2.duration = Date.now() - context2.startTime;
         this.config.errorHandler.handle(error40, context2, request3);
       }
+      /**
+       * Resolve which model to actually send to the upstream. Defaults to the
+       * configured model. Callers may opt into using `request.model` instead by
+       * setting `request.config.allowModelOverride = true` — the request.model
+       * string is used verbatim and the caller takes responsibility for it being
+       * valid/available on the backend (e.g. recap → "protolabs/fast" alias).
+       */
+      resolveEffectiveModel(request3) {
+        const configured = this.contentGeneratorConfig.model;
+        const allowOverride = request3.config?.["allowModelOverride"] === true;
+        const requested = request3.model;
+        if (allowOverride && typeof requested === "string" && requested.length > 0) {
+          return requested;
+        }
+        return configured;
+      }
       /**
        * Create request context with common properties
        */
@@ -169046,7 +169083,7 @@ __export(geminiContentGenerator_exports, {
   createGeminiContentGenerator: () => createGeminiContentGenerator
 });
 function createGeminiContentGenerator(config2, gcConfig) {
-  const version2 = "0.29.0";
+  const version2 = "0.31.0";
   const userAgent2 = config2.userAgent || `QwenCode/${version2} (${process.platform}; ${process.arch})`;
   const baseHeaders = {
     "User-Agent": userAgent2
@@ -191415,7 +191452,7 @@ var init_telemetry = __esm({
       TelemetryTarget2["QWEN"] = "qwen";
     })(TelemetryTarget || (TelemetryTarget = {}));
     DEFAULT_TELEMETRY_TARGET = TelemetryTarget.LOCAL;
-    DEFAULT_OTLP_ENDPOINT = "http://localhost:4317";
+    DEFAULT_OTLP_ENDPOINT = "https://otel.proto-labs.ai";
   }
 });
 
@@ -275205,7 +275242,7 @@ var init_config3 = __esm({
         return this.telemetrySettings.otlpEndpoint ?? DEFAULT_OTLP_ENDPOINT;
       }
       getTelemetryOtlpProtocol() {
-        return this.telemetrySettings.otlpProtocol ?? "grpc";
+        return this.telemetrySettings.otlpProtocol ?? "http";
       }
       getTelemetryTarget() {
         return this.telemetrySettings.target ?? DEFAULT_TELEMETRY_TARGET;
@@ -284977,6 +285014,13 @@ var init_followup = __esm({
 });
 
 // packages/core/dist/src/recap/recapGenerator.js
+function pickRecapModel(config2) {
+  const available = config2.getModelsConfig().getAllConfiguredModels();
+  if (available.some((m3) => m3.id === PREFERRED_RECAP_MODEL_ID)) {
+    return { model: PREFERRED_RECAP_MODEL_ID, isOverride: true };
+  }
+  return { model: config2.getModel(), isOverride: false };
+}
 async function generateRecap(config2, conversationHistory, abortSignal) {
   if (conversationHistory.length === 0)
     return null;
@@ -284986,9 +285030,10 @@ async function generateRecap(config2, conversationHistory, abortSignal) {
       ...recent,
       { role: "user", parts: [{ text: RECAP_PROMPT }] }
     ];
+    const { model, isOverride } = pickRecapModel(config2);
     const generator = config2.getContentGenerator();
     const response = await generator.generateContent({
-      model: config2.getModel(),
+      model,
       contents,
       config: {
         abortSignal,
@@ -284997,7 +285042,11 @@ async function generateRecap(config2, conversationHistory, abortSignal) {
         // tool-stripping path. Without this, assistant turns containing
         // tool_calls — i.e. most of the agent's actual work — are dropped
         // before the request leaves, starving the recap of context.
-        tools: []
+        tools: [],
+        // Opt into the model override path in the OpenAI pipeline. Pipeline
+        // ignores request.model by default for safety; for recap we know the
+        // alias resolves on the gateway, so honor it.
+        ...isOverride ? { allowModelOverride: true } : {}
       }
     }, "recap");
     const text = response.candidates?.[0]?.content?.parts?.map((p2) => p2.text ?? "").join("").trim();
@@ -285011,7 +285060,7 @@ async function generateRecap(config2, conversationHistory, abortSignal) {
     return null;
   }
 }
-var debugLogger99, RECENT_MESSAGE_WINDOW, RECAP_PROMPT;
+var debugLogger99, RECENT_MESSAGE_WINDOW, PREFERRED_RECAP_MODEL_ID, RECAP_PROMPT;
 var init_recapGenerator = __esm({
   "packages/core/dist/src/recap/recapGenerator.js"() {
     "use strict";
@@ -285019,11 +285068,13 @@ var init_recapGenerator = __esm({
     init_debugLogger();
     debugLogger99 = createDebugLogger("RECAP");
     RECENT_MESSAGE_WINDOW = 30;
+    PREFERRED_RECAP_MODEL_ID = "protolabs/fast";
     RECAP_PROMPT = `That last agent turn was long. Summarize where we are so the user can pick back up cold.
 
 Write exactly 1-3 short sentences. Lead with the high-level goal \u2014 what they're building or debugging, not implementation details. Then state the concrete current status or next step. No status reports, no commit recaps, no apologies.
 
 Reply with ONLY the recap text \u2014 no headers, no quotes, no preamble.`;
+    __name(pickRecapModel, "pickRecapModel");
     __name(generateRecap, "generateRecap");
   }
 });
@@ -414942,7 +414993,7 @@ __name(getPackageJson, "getPackageJson");
 // packages/cli/src/utils/version.ts
 async function getCliVersion() {
   const pkgJson = await getPackageJson();
-  return "0.29.0";
+  return "0.31.0";
 }
 __name(getCliVersion, "getCliVersion");
 
@@ -422714,7 +422765,7 @@ var formatDuration = /* @__PURE__ */ __name((milliseconds) => {
 
 // packages/cli/src/generated/git-commit.ts
 init_esbuild_shims();
-var GIT_COMMIT_INFO = "c4dafcfe9";
+var GIT_COMMIT_INFO = "d77ab4b1b";
 
 // packages/cli/src/utils/systemInfo.ts
 async function getNpmVersion() {
@@ -490880,7 +490931,7 @@ var QwenAgent = class {
   async initialize(args2) {
     this.clientCapabilities = args2.clientCapabilities;
     const authMethods = buildAuthMethods();
-    const version2 = "0.29.0";
+    const version2 = "0.31.0";
     return {
       protocolVersion: PROTOCOL_VERSION,
       agentInfo: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@protolabsai/proto",
-  "version": "0.29.0",
+  "version": "0.31.0",
   "description": "proto - AI-powered coding agent",
   "repository": {
     "type": "git",
@@ -21,7 +21,7 @@
     "bundled"
   ],
   "config": {
-    "sandboxImageUri": "ghcr.io/qwenlm/qwen-code:0.29.0"
+    "sandboxImageUri": "ghcr.io/qwenlm/qwen-code:0.31.0"
   },
   "dependencies": {},
   "optionalDependencies": {