> ## Documentation Index
> Fetch the complete documentation index at: https://docs.valiqor.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Ollama

> Auto-instrument Ollama local model calls with tracing for chat and generate endpoints.

Valiqor automatically traces Ollama API calls by intercepting HTTP
requests to the Ollama server. Both `/api/chat` and `/api/generate`
endpoints are captured with model name, token usage, and response content.

<Info>
  Ollama tracing works by detecting HTTP requests to the Ollama API server.
  No additional install extras are needed — Ollama support is built into the
  core `valiqor` package.
</Info>

***

## Install

```bash theme={"system"}
pip install valiqor
```

No extra dependencies are needed. Ollama tracing uses the `requests`
library which is already a core dependency.

Make sure you have [Ollama](https://ollama.com) installed and running
locally with a model pulled:

```bash theme={"system"}
ollama pull llama3.2
```

***

## Zero-Config (Recommended)

Add a single import at the top of your app — all Ollama calls are
automatically traced:

```python theme={"system"}
import valiqor.auto  # ← Add this line

import requests

# Chat endpoint
response = requests.post("http://localhost:11434/api/chat", json={
    "model": "llama3.2",
    "messages": [{"role": "user", "content": "Explain quantum computing"}],
    "stream": False
})
print(response.json()["message"]["content"])
```

***

## Selective Instrumentation

If you only want Ollama tracing:

```python theme={"system"}
from valiqor.trace import ollama_autolog

ollama_autolog()
```

***

## Chat Endpoint

The `/api/chat` endpoint uses the messages format:

```python theme={"system"}
import valiqor.auto
import requests

response = requests.post("http://localhost:11434/api/chat", json={
    "model": "llama3.2",
    "messages": [
        {"role": "system", "content": "You are a helpful assistant"},
        {"role": "user", "content": "What is Python?"}
    ],
    "stream": False
})

result = response.json()
print(result["message"]["content"])
```

***

## Generate Endpoint

The `/api/generate` endpoint uses a prompt string:

```python theme={"system"}
import valiqor.auto
import requests

response = requests.post("http://localhost:11434/api/generate", json={
    "model": "llama3.2",
    "prompt": "Explain machine learning in simple terms",
    "stream": False
})

result = response.json()
print(result["response"])
```

***

## What Gets Captured

Each traced Ollama call records:

| Field               | Description                                  |
| ------------------- | -------------------------------------------- |
| `model`             | Model name (e.g. `llama3.2`, `mistral`)      |
| `endpoint`          | API endpoint (`chat` or `generate`)          |
| `prompt_tokens`     | Input token count (from `prompt_eval_count`) |
| `completion_tokens` | Output token count (from `eval_count`)       |
| `messages`          | User prompt or messages                      |
| `response`          | Model response text                          |
| `duration_ms`       | Call latency                                 |
| `status`            | Success or error                             |

***

## With Workflows

Group multiple Ollama calls into a single trace:

```python theme={"system"}
import valiqor.auto
from valiqor.trace import trace_workflow
import requests

OLLAMA_URL = "http://localhost:11434/api/chat"

with trace_workflow("local-assistant"):
    # Step 1: Research
    research = requests.post(OLLAMA_URL, json={
        "model": "llama3.2",
        "messages": [{"role": "user", "content": "What are the key facts about Mars?"}],
        "stream": False
    })

    # Step 2: Summarize
    summary = requests.post(OLLAMA_URL, json={
        "model": "llama3.2",
        "messages": [
            {"role": "user", "content": f"Summarize: {research.json()['message']['content']}"}
        ],
        "stream": False
    })
```

***

## Disabling

```python theme={"system"}
from valiqor.trace import disable_autolog

disable_autolog("ollama")    # Disable Ollama only
disable_autolog()            # Disable all providers
```

***

## Limitations

* **Sync only** — only synchronous `requests.post` calls are traced.
  Async HTTP clients (e.g. `httpx`, `aiohttp`) are not intercepted.
* **Streaming** is not instrumented — set `"stream": False` in your
  requests for traces to be captured.
* **Ollama Python library** — if you use the `ollama` Python package
  instead of raw `requests`, it works as long as it uses `requests.post`
  internally.

***

## Next Steps

<CardGroup cols={2}>
  <Card title="Tracing Guide" icon="route" href="/workflows/tracing">
    Learn about traces, spans, workflows, and exporters.
  </Card>

  <Card title="Failure Analysis" icon="magnifying-glass" href="/workflows/failure-analysis">
    Run failure analysis on your traced Ollama calls.
  </Card>
</CardGroup>