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

# Environment variables

> What Client::from_env() reads and which env vars Cognis honors at runtime.

`Client::from_env()` reads a small set of environment variables to pick a provider, a key, and a default model. There are also a handful of runtime variables that affect behavior at a lower level. None are required if you build clients explicitly via `Client::builder()`.

## Provider selection

| Variable          | Required             | Notes                                                                    |
| ----------------- | -------------------- | ------------------------------------------------------------------------ |
| `COGNIS_PROVIDER` | yes (for `from_env`) | One of `openai`, `anthropic`, `google`, `ollama`, `azure`, `openrouter`. |

## Per-provider credentials and defaults

Substitute `<PROVIDER>` with `OPENAI`, `ANTHROPIC`, `GOOGLE`, `OLLAMA`, `AZURE`, or `OPENROUTER`.

| Variable                     | Default          | Notes                                                              |
| ---------------------------- | ---------------- | ------------------------------------------------------------------ |
| `COGNIS_<PROVIDER>_API_KEY`  | —                | Provider key. Not needed for Ollama.                               |
| `COGNIS_<PROVIDER>_MODEL`    | provider default | Default model name. Override per call via `ChatOptions::model`.    |
| `COGNIS_<PROVIDER>_BASE_URL` | provider default | Override the API base. Useful for self-hosted or proxied backends. |

Provider defaults vary; check the provider builder's source for the exact value if you don't override.

### Azure-specific

| Variable                   | Notes                                                        |
| -------------------------- | ------------------------------------------------------------ |
| `COGNIS_AZURE_API_KEY`     | Azure OpenAI key.                                            |
| `COGNIS_AZURE_ENDPOINT`    | E.g. `https://your-resource.openai.azure.com`.               |
| `COGNIS_AZURE_DEPLOYMENT`  | The deployment name (your custom alias).                     |
| `COGNIS_AZURE_API_VERSION` | E.g. `2024-08-06`.                                           |
| `COGNIS_AZURE_MODEL`       | Optional model name (Azure typically uses deployment names). |

## Langfuse (cognis-trace, feature `langfuse`)

`LangfuseExporter::from_env()` and `LangfuseConfig::from_env()` read these. *Not* prefixed with `COGNIS_` to match Langfuse's conventions.

| Variable              | Required | Notes                                                               |
| --------------------- | -------- | ------------------------------------------------------------------- |
| `LANGFUSE_PUBLIC_KEY` | yes      | `pk-lf-...`                                                         |
| `LANGFUSE_SECRET_KEY` | yes      | `sk-lf-...`                                                         |
| `LANGFUSE_HOST`       | optional | Defaults to `https://cloud.langfuse.com`. Override for self-hosted. |

## Examples

```bash theme={null}
# OpenAI:
export COGNIS_PROVIDER=openai
export COGNIS_OPENAI_API_KEY=sk-...
export COGNIS_OPENAI_MODEL=gpt-4o-mini

# Anthropic:
export COGNIS_PROVIDER=anthropic
export COGNIS_ANTHROPIC_API_KEY=sk-ant-...

# Local Ollama:
export COGNIS_PROVIDER=ollama
export COGNIS_OLLAMA_MODEL=llama3.1
# COGNIS_OLLAMA_BASE_URL defaults to http://localhost:11434

# Azure:
export COGNIS_PROVIDER=azure
export COGNIS_AZURE_API_KEY=...
export COGNIS_AZURE_ENDPOINT=https://your-resource.openai.azure.com
export COGNIS_AZURE_DEPLOYMENT=gpt-4o-prod
export COGNIS_AZURE_API_VERSION=2024-08-06

# OpenRouter:
export COGNIS_PROVIDER=openrouter
export COGNIS_OPENROUTER_API_KEY=...
export COGNIS_OPENROUTER_MODEL=anthropic/claude-sonnet-4
```

## Runtime variables

Cognis honors the standard `tracing` env vars for log filtering:

```bash theme={null}
export RUST_LOG=cognis=info,cognis_trace=debug
```

There is no global verbosity flag; control it through `tracing-subscriber`'s filter directives.

## What about `.env` files?

Cognis **does not read `.env` files**. Plain `.env` on disk is a security footgun (they ship to backups, logs, and screen-sharing surfaces). Recommended setup on macOS / Linux: `direnv` + `envchain` so the same shell-loaded vars are picked up automatically. See [Installation → Set credentials](/get-started/installation#set-credentials).

For CI: set the same variables in your runner's secret store.

## See also

<CardGroup cols={2}>
  <Card title="Installation" icon="download" href="/get-started/installation">
    The full credentials setup.
  </Card>

  <Card title="Models and providers" icon="plug" href="/building-agents/models">
    Builder shapes when env vars aren't enough.
  </Card>
</CardGroup>