Add guide for tracing and logging Roo Code API POST requests with Braintrust AI Proxy

Author: d-oit

adminEditor/data/categories.json

@@ -6,7 +6,7 @@
     },
     {
       "name": "Software Development",
-      "count": 5
+      "count": 6
     }
   ]
 }

adminEditor/data/tags.json

@@ -43,6 +43,14 @@
     {
       "name": "GitHub",
       "count": 1
+    },
+    {
+      "name": "AI",
+      "count": 1
+    },
+    {
+      "name": "Roo Code",
+      "count": 1
     }
   ]
 }

assets/img/blog/RooCodeBraintrustTrace/BraintrustProjectId.png

@@ -0,0 +1,87 @@
+---
+title: "Trace API POST Requests: Roo Code + Braintrust AI Guide"
+slug: how-to-trace-and-log-roo-code-api-post-requests-with-braintrust-ai-proxy
+description: Learn how to configure Roo Code with Braintrust AI Proxy for enhanced API request tracing, caching, and OpenTelemetry monitoring - a step-by-step implementation guide.
+date: 2025-05-19T09:55:00Z
+tags:
+   - AI
+   - Roo Code
+categories:
+   - Software Development
+thumbnail:
+   url: /img/blog/how-to-trace-and-log-roo-code-api-post-requests-with-braintrust-ai-proxy.jpg
+   author: d.o. (flux)
+draft: false
+---
+
+Here’s how to configure **Roo Code**’s UI to route all your OpenAI–compatible requests through the Braintrust AI Proxy—complete with transparent caching, built-in tracing, and full OpenTelemetry support—without writing a single line of code. By simply adjusting the **API Provider**, **Base URL**, and **Custom Headers** fields in your Roo Code settings panel (as shown in your screenshots), you unlock:
+
+* **Cache modes** (`auto`, `always`, `never`) that let you reuse deterministic outputs when `temperature=0`
+* **Tracing headers** (`x-bt-parent: project_id:<YOUR_PROJECT_ID>`) that generate root spans in Braintrust’s monitoring dashboard
+* **OpenTelemetry export** via OTLP endpoints, compatible with any observability platform (including Traceloop’s OpenLLMetry)
+
+## Configuring Roo Code’s UI
+
+1. **API Provider**
+
+   * Select **OpenAI Compatible** from the dropdown. This tells Roo Code to treat the endpoint exactly like the OpenAI API, so you can continue using the same SDK calls internally.
+
+2. **Base URL**: https://api.braintrust.dev/v1/proxy
+
+   * This single endpoint proxies across multiple model providers while adding caching and observability. [List of supported models and providers](https://www.braintrust.dev/docs/guides/proxy?utm_source=chatgpt.com#list-of-supported-models-and-providers)
+
+3. **OpenAI API Key**
+
+   * Paste your **Braintrust** API key here.
+   * Whenever you include tracing headers, Braintrust uses this key to authenticate and associate spans with your project.
+
+   {{< image class="rounded lightbox" src="/img/blog/RooCodeBraintrustTrace/RooCodeBraintrustTraceCacheConfiguration.png" caption="Click the image to enlarge and view the full configuration: Insert the Braintrust proxy endpoint into Roo Code's Base URL field. Add the Braintrust-specific headers to enable tracing and caching." >}}
+
+4. **Custom Headers**
+
+   * Click **+** to add headers exactly as shown in your UI:
+
+     * `x-bt-use-cache` = `auto`
+     * `x-bt-parent` = `project_id:<YOUR_PROJECT_ID>`
+   * These instruct the proxy to cache deterministic calls and to start a new trace under your Braintrust project.
+
+   {{< image class="rounded lightbox" src="/img/blog/RooCodeBraintrustTrace/BraintrustProjectId.png" caption="Get project id in braintrust web ui." >}}
+
+5. **Enable Streaming** (optional)
+
+   * Check **Enable streaming** if you want to stream partial responses back to Roo Code’s editor.
+
+6. **Other toggles**
+
+   * Leave **Use Azure** and **Enable R1 parameters** unchecked unless explicitly needed by your custom provider.
+
+---
+
+## How Caching Works
+
+Braintrust supports three cache modes set via `x-bt-use-cache`:
+
+* **`auto`**: caches only when `temperature=0` or a deterministic `seed` is provided (the default mode)
+* **`always`**: caches every supported request (e.g. `/chat/completions`, `/completions`, `/embeddings`) regardless of randomness settings
+* **`never`**: bypasses cache entirely—useful for debugging or non-deterministic experiments
+
+> **Pro tip:** To get a cache hit, set **Temperature** to **0** in your Roo Code run configuration. This makes the output deterministic and eligible for replay.
+
+---
+
+## How Tracing Works
+
+Every request that includes `x-bt-parent` will emit a root span into Braintrust’s tracing backend:
+
+* **Header**: `x-bt-parent: project_id:<YOUR_PROJECT_ID>`
+* **Key**: your **Braintrust API Key** in the **OpenAI API Key** field
+* **Outcome**: each API call is logged as an LLM span under your project, visible in the Braintrust dashboard
+
+This gives you full visibility into prompt inputs, response tokens, timings, and cache hits—all without modifying any application code.
+
+{{< image class="rounded lightbox" src="/img/blog/RooCodeBraintrustTrace/BraintrustRooCodeChatTrace.png" caption="Braintrust chat trace" >}}
+
+## Further Reading
+
+* **Braintrust AI Proxy Guide** (caching, headers, endpoints) {{< link "https://github.com/braintrustdata/braintrust-proxy" >}}Braintrust AI proxy{{< /link >}}
+* **Traceloop OpenLLMetry** (semantic conventions for LLM) {{< link "https://github.com/traceloop/openllmetry" >}}Traceloop OpenLLMetry{{< /link >}}

assets/img/blog/RooCodeBraintrustTrace/BraintrustRooCodeChatTrace.png

@@ -13,7 +13,6 @@
       "figure",
       "footer",
       "form",
-      "h1",
       "h2",
       "h3",
       "h4",
@@ -525,6 +524,7 @@
       "code-verwendung",
       "conclusion",
       "configuration-front-matter-output",
+      "configuring-roo-codes-ui",
       "contact",
       "content",
       "cookies",
@@ -605,6 +605,7 @@
       "free-basket-streams-table",
       "frontend",
       "fullscreenandprintbuttonhtml",
+      "further-reading",
       "gallery",
       "generate-a-starting-point-for-all-websites-pages",
       "generated-playwright-js-file",
@@ -621,7 +622,9 @@
       "hinode-module-a-slideshow-gallery-shortcode-for-hugo-compatible-with-hinode",
       "history",
       "hosting",
+      "how-caching-works",
       "how-to-use-mistral-api-with-cline-or-cline-fork",
+      "how-tracing-works",
       "html-theme-toggle-interface",
       "hugo",
       "hugo-module-aktualisieren",
@@ -643,7 +646,6 @@
       "list-of-modified-files",
       "listView",
       "liveToastSaveThemMode",
-      "log-llm-api-post-from-roo-code",
       "lokale-installation",
       "my-text-image-generation-for-different-models",
       "navbar-0-collapse",