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

# Claude Code CLI

> Point Anthropic's Claude Code CLI at ByteSpike. Two env vars, done — works against every Anthropic-compatible model in the catalog.

Claude Code is Anthropic's coding agent CLI. Out of the box it talks
to Anthropic's gateway; redirecting it at ByteSpike unlocks the rest
of the model catalog (Claude, plus DeepSeek-Anthropic, plus
Anthropic-via-Gemini translation, etc.) under one key.

## Prerequisites

* A ByteSpike account + a key bound to the **`claude-default`** group
  (or any group that includes Claude/DeepSeek-Anthropic). See
  [Register](/register).
* Claude Code installed:
  ```bash theme={null}
  npm install -g @anthropic-ai/claude-code
  # or
  brew install claude-code
  ```

## Configure

Two environment variables. Add to your shell profile (`~/.zshrc`,
`~/.bashrc`, or `~/.config/fish/config.fish`):

```bash theme={null}
export ANTHROPIC_BASE_URL="https://llm.bytespike.ai"
export ANTHROPIC_API_KEY="sk-byts-..."
```

Reload the shell (`exec $SHELL` or open a new tab). That's it —
`claude` now sends every Messages-API call through ByteSpike.

<Tip>
  Want it scoped to one project, not your whole shell? Drop a
  `.envrc` (with `direnv`) or a project-local script that exports
  the vars only inside the repo:

  ```bash theme={null}
  # .envrc
  export ANTHROPIC_BASE_URL="https://llm.bytespike.ai"
  export ANTHROPIC_API_KEY="sk-byts-..."
  ```
</Tip>

## Verify

```bash theme={null}
claude --version
```

Then start a session:

```bash theme={null}
claude
```

In the running session, try a small task:

```
> what model are you?
```

You should see the model self-identify (e.g. "I'm Claude Sonnet
4.6 served via ByteSpike"). If you get an auth error,
double-check `ANTHROPIC_API_KEY`; if you get `400 model_not_allowed`,
your key's group doesn't include the model Claude Code defaulted to
— switch to a group that includes it or override with `--model`.

## Switching models

Claude Code picks the right tool per command. Override per-session:

```bash theme={null}
claude --model claude-opus-4-8    # use Opus for this session
```

Or per-tool inside the session:

```
> /model claude-haiku-4-5
```

Any model id from your `claude-default` group (or whatever group
your key is bound to) works. See [Models](/models) for the full
catalog.

## Common configs

<AccordionGroup>
  <Accordion title="One key per project">
    Generate a separate key per project from
    [Console → API keys](https://console.bytespike.ai/keys) with its
    own `quota` (USD lifetime cap) and `rate_limit_*` buckets. Wire
    each project's `.envrc` to the right key. Spend stays
    attributable in [`/api/v1/me/usage`](/api-reference/account/me-usage)
    by `api_key_id`.
  </Accordion>

  <Accordion title="Streaming + tool use">
    Both work end-to-end through ByteSpike — no extra config. The
    gateway preserves `tool_use` blocks, `cache_control` markers, and
    `thinking` blocks (Opus / Sonnet 4.x) untouched.
  </Accordion>

  <Accordion title="MCP server compatibility">
    Claude Code's MCP servers (`mcp__*` tools) all run client-side —
    they don't care which gateway sits between the CLI and the model.
    Just set the env vars and your existing MCP config keeps working.
  </Accordion>

  <Accordion title="Behind a corporate proxy">
    Claude Code respects `HTTPS_PROXY` / `NO_PROXY`. Add
    `llm.bytespike.ai` to `NO_PROXY` if your proxy mangles SSE
    streams; otherwise set `HTTPS_PROXY=https://proxy.corp:443` and
    proceed normally.
  </Accordion>
</AccordionGroup>

## Troubleshooting

| Symptom                         | Cause                                                         | Fix                                                                                                                                           |
| ------------------------------- | ------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------- |
| `401 invalid x-api-key`         | Wrong key value or key was revoked                            | Re-copy from [Console → API keys](https://console.bytespike.ai/keys); rotate if needed                                                        |
| `402 insufficient_balance`      | Account out of credits or key hit its `quota`                 | Top up at [Console → Billing](https://console.bytespike.ai/billing); raise the key's quota                                                    |
| `403 permission_denied` (model) | Model not in the key's routing group                          | Switch key, or switch model with `--model`                                                                                                    |
| `503 api_error`                 | The key's group has no capacity to serve this model right now | Click **Test** next to the model in [Console → Models](https://console.bytespike.ai/dosia/models) to confirm; switch group if it's a mismatch |
| Mid-stream stall (SSE)          | Corporate proxy buffering                                     | Add `llm.bytespike.ai` to `NO_PROXY`                                                                                                          |

For the live response shape, see
[`/v1/messages`](/api-reference/text/claude-messages).

## Next

<CardGroup cols={2}>
  <Card title="Codex CLI" icon="terminal" href="/codex-cli">
    OpenAI's coding CLI — same idea, OpenAI-side env vars.
  </Card>

  <Card title="Models" icon="boxes-stacked" href="/models">
    What Claude Code can actually call.
  </Card>

  <Card title="Authentication" icon="key" href="/authentication">
    Group binding, quota, rate limits.
  </Card>

  <Card title="DOSIA Agent mode" icon="robot" href="/concepts/dosia-agent-mode">
    The agent integration that builds on top of Claude Code's Messages shape.
  </Card>
</CardGroup>