diff --git a/python/cerebrium/.env.example b/python/cerebrium/.env.example
new file mode 100644
index 0000000..8727610
--- /dev/null
+++ b/python/cerebrium/.env.example
@@ -0,0 +1,21 @@
+# Cerebrium secrets — set these in the Cerebrium dashboard under
+# Project Settings → Secrets, or with `cerebrium secrets`. They surface
+# inside the container as environment variables.
+
+# Last9 OTLP HTTP endpoint base — see Integrations → OpenTelemetry in app.last9.io
+OTEL_EXPORTER_OTLP_ENDPOINT=https://otlp-aps1.last9.io
+
+# Last9 auth header in OTEL env-var format (note the inner '='):
+#   <header-name>=<header-value>
+# Copy the full Basic value from Last9's OpenTelemetry integration page.
+OTEL_EXPORTER_OTLP_HEADERS=Authorization=Basic <your-last9-basic-token>
+
+# OTLP over HTTP/protobuf (4318 path semantics, not 4317 gRPC).
+OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
+
+# Optional: override what shows up as service.name in Last9. Defaults
+# to the value of CEREBRIUM_APP_NAME if present.
+OTEL_SERVICE_NAME=otel-autoinstrument-last9
+
+# Optional: surface deployment.environment on every span/metric.
+OTEL_RESOURCE_ATTRIBUTES=deployment.environment=production
diff --git a/python/cerebrium/.gitignore b/python/cerebrium/.gitignore
new file mode 100644
index 0000000..8ee566b
--- /dev/null
+++ b/python/cerebrium/.gitignore
@@ -0,0 +1,27 @@
+# Environment/secrets
+.env
+.env.local
+.env.*.local
+
+# Cerebrium build artifacts
+.cerebrium/
+build/
+
+# Python
+__pycache__/
+*.pyc
+*.pyo
+.venv/
+venv/
+
+# IDE
+.idea/
+.vscode/
+*.swp
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Logs
+*.log
diff --git a/python/cerebrium/README.md b/python/cerebrium/README.md
new file mode 100644
index 0000000..fd0cacb
--- /dev/null
+++ b/python/cerebrium/README.md
@@ -0,0 +1,83 @@
+# Cerebrium + OpenTelemetry Auto-Instrumentation → Last9
+
+Deploy a FastAPI app on Cerebrium with zero OpenTelemetry code in your source files. `opentelemetry-instrument` runs in the entrypoint and auto-patches FastAPI, `requests`, `httpx`, and logging at process startup. Traces ship to Last9 over OTLP/HTTP.
+
+This complements Cerebrium's platform-side **metrics export** (resource and execution metrics shipped from the platform itself). Together you get metrics + traces in Last9.
+
+## Prerequisites
+
+- Cerebrium account with the CLI installed: `pip install cerebrium`
+- Last9 account — grab the OTLP HTTP endpoint and Auth Header from **Integrations → OpenTelemetry** in [app.last9.io](https://app.last9.io)
+
+## Quick Start
+
+1. **Clone this directory and set up secrets in Cerebrium.**
+
+   ```bash
+   cerebrium login
+   cerebrium secrets set OTEL_EXPORTER_OTLP_ENDPOINT https://otlp-aps1.last9.io
+   cerebrium secrets set OTEL_EXPORTER_OTLP_HEADERS 'Authorization=Basic <your-token>'
+   cerebrium secrets set OTEL_EXPORTER_OTLP_PROTOCOL http/protobuf
+   cerebrium secrets set OTEL_SERVICE_NAME otel-autoinstrument-last9
+   ```
+
+   See `.env.example` for the full list of variables.
+
+2. **Deploy.**
+
+   ```bash
+   cerebrium deploy
+   ```
+
+   Build runs `opentelemetry-bootstrap --action=install` which detects every supported library in your pip deps and pulls the matching instrumentation package. Look for it in the build logs.
+
+3. **Invoke the app.**
+
+   ```bash
+   curl -X POST https://api.cortex.cerebrium.ai/v4/p-XXXXXX/otel-autoinstrument-last9/predict \
+     -H "Authorization: Bearer <cerebrium-token>" \
+     -H "Content-Type: application/json" \
+     -d '{"prompt": "trace me", "run_id": "test-1"}'
+   ```
+
+4. **View traces in Last9.**
+
+   Open [Traces Explorer](https://app.last9.io/traces), filter by `service.name = otel-autoinstrument-last9`. You should see a `POST /predict` parent span with child `GET` spans for the two `httpbin.org` calls.
+
+## How it works
+
+- `cerebrium.toml` defines `entrypoint = ["opentelemetry-instrument", "uvicorn", "main:app", ...]`. Cerebrium runs that command directly, so the OTel launcher gets first-class control of process startup.
+- `opentelemetry-instrument` reads `OTEL_*` env vars, sets `PYTHONPATH` to OTel's `sitecustomize.py`, then `exec`s your command. From that point every Python import gets the OTel-patched version of the library.
+- `main.py` is just FastAPI. Nothing else.
+
+## What gets auto-instrumented
+
+Anything in the dependencies that has an `opentelemetry-instrumentation-*` package on PyPI. Common picks for AI workloads:
+
+| Library | Captures |
+|---|---|
+| `fastapi` | Request span per route, status code, route template |
+| `requests`, `httpx`, `urllib3` | Outbound HTTP spans with method, URL, status |
+| `logging` | Trace and span IDs injected into log records |
+| `openai`, `anthropic` | Per-call LLM spans with model, token counts |
+| `sqlalchemy`, `psycopg2`, `redis` | DB and cache spans |
+
+To add more, list them in `[cerebrium.dependencies.pip]` and re-deploy — `opentelemetry-bootstrap` picks them up.
+
+## Combining with Cerebrium's metrics export
+
+This example handles **traces only**. For platform-side **metrics** (CPU, GPU, container counts, run latencies), configure Cerebrium's dashboard:
+
+- **Integrations → Metrics Export → Custom OTLP**
+- Endpoint: `https://otlp-aps1.last9.io` (base URL — Cerebrium auto-appends `/v1/metrics`)
+- Auth Header Name: `Authorization`
+- Auth Header Value: `Basic <token>`
+
+Full guide: [Cerebrium integration on docs.last9.io](https://last9.io/docs/integrations/cerebrium)
+
+## Troubleshooting
+
+- **No traces in Last9.** Check Cerebrium build logs for the `opentelemetry-bootstrap` output — it must list at least `fastapi` and `requests`. If absent, the deps didn't install or the shell command failed.
+- **`AttributeError: ... has no attribute 'instrument'` at startup.** A library was imported before `opentelemetry-instrument` could patch it — make sure no `pre_build_commands` or custom shell commands import your app modules.
+- **Header malformed errors from OTLP.** `OTEL_EXPORTER_OTLP_HEADERS` uses `=` between header name and value, not `:`. Correct format: `Authorization=Basic <token>`.
+- **Spans appear but with no parent/child link.** A library is being imported at module top level before the sitecustomize fires. Move the import inside the function, or add an explicit `*Instrumentor().instrument()` call in a small bootstrap module loaded first.
diff --git a/python/cerebrium/cerebrium.toml b/python/cerebrium/cerebrium.toml
new file mode 100644
index 0000000..31911a9
--- /dev/null
+++ b/python/cerebrium/cerebrium.toml
@@ -0,0 +1,51 @@
+# Cerebrium app with OpenTelemetry auto-instrumentation exporting to Last9.
+# main.py contains zero OTel code — opentelemetry-instrument in the entrypoint
+# does all the patching at process startup.
+
+[cerebrium.deployment]
+name = "otel-autoinstrument-last9"
+python_version = "3.12"
+include = ["*"]
+exclude = [".*", "__pycache__", "*.pyc"]
+
+# Install all available instrumentors after pip deps are resolved.
+# opentelemetry-bootstrap detects every supported library in the env and
+# pins the matching instrumentation package for opentelemetry-instrument.
+shell_commands = [
+  "opentelemetry-bootstrap --action=install",
+]
+
+[cerebrium.hardware]
+cpu = 1
+memory = 2
+compute = "CPU"
+
+[cerebrium.scaling]
+min_replicas = 0
+max_replicas = 1
+cooldown = 60
+
+[cerebrium.dependencies.pip]
+fastapi = "*"
+uvicorn = "*"
+requests = "*"
+httpx = "*"
+# Pin OTel core + semconv together. Without this, pip can resolve a newer
+# exporter-otlp-proto-common (which imports semconv._incubating) against an
+# older semantic-conventions wheel — startup fails with ModuleNotFoundError.
+opentelemetry-distro = ">=0.51b0"
+opentelemetry-exporter-otlp = ">=1.30.0"
+opentelemetry-semantic-conventions = ">=0.51b0"
+
+[cerebrium.runtime.custom]
+port = 8000
+healthcheck_endpoint = "/health"
+readycheck_endpoint = "/ready"
+# Prepending opentelemetry-instrument is the auto-instrumentation hook.
+# It sets PYTHONPATH to the OTel sitecustomize, then execs uvicorn.
+entrypoint = [
+  "opentelemetry-instrument",
+  "uvicorn", "main:app",
+  "--host", "0.0.0.0",
+  "--port", "8000",
+]
diff --git a/python/cerebrium/main.py b/python/cerebrium/main.py
new file mode 100644
index 0000000..a093182
--- /dev/null
+++ b/python/cerebrium/main.py
@@ -0,0 +1,53 @@
+"""Cerebrium app — zero OpenTelemetry code.
+
+All tracing is wired by `opentelemetry-instrument` in the entrypoint
+(see cerebrium.toml). FastAPI, requests, and httpx are auto-patched.
+"""
+
+import logging
+import time
+
+import requests
+from fastapi import FastAPI
+from pydantic import BaseModel
+
+logging.basicConfig(level=logging.INFO)
+log = logging.getLogger(__name__)
+
+app = FastAPI()
+
+
+class PredictRequest(BaseModel):
+    prompt: str = "hello world"
+    run_id: str | None = None
+
+
+@app.get("/health")
+def health():
+    return {"status": "ok"}
+
+
+@app.get("/ready")
+def ready():
+    return {"status": "ok"}
+
+
+@app.post("/predict")
+def predict(req: PredictRequest):
+    log.info("predict: prompt_len=%d run_id=%s", len(req.prompt), req.run_id)
+    start = time.perf_counter()
+    statuses = []
+    for url in ("https://httpbin.org/uuid", "https://httpbin.org/headers"):
+        try:
+            r = requests.get(url, timeout=5)
+            statuses.append(r.status_code)
+        except Exception as e:
+            log.warning("downstream call failed: url=%s err=%s", url, e)
+            statuses.append(-1)
+    latency_ms = (time.perf_counter() - start) * 1000.0
+    log.info("predict: done in %.1fms statuses=%s", latency_ms, statuses)
+    return {
+        "prompt": req.prompt,
+        "downstream_status": statuses,
+        "latency_ms": round(latency_ms, 2),
+    }