> ## 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.
# Migrate from Anthropic
> Switch from Anthropic's API to ByteSpike with two env vars. Same Anthropic SDK, same Messages shape, every Claude + cross-vendor model in our catalog.
If you're already on Anthropic's API, switching to ByteSpike means
**changing two env vars and shipping**. The Anthropic SDK works
unchanged; the request shape is identical; only the base URL and
key differ.
## The two env vars
```diff theme={null}
- export ANTHROPIC_API_KEY=sk-ant-...
+ export ANTHROPIC_BASE_URL=https://llm.bytespike.ai
+ export ANTHROPIC_API_KEY=sk-byts-...
```
Most clients (Claude Code, Claude Desktop, Anthropic SDKs Python /
TypeScript, third-party tools) read these standard vars
automatically — no code change.
If you construct the client explicitly:
```diff theme={null}
- client = Anthropic() # reads env vars
+ client = Anthropic(
+ base_url="https://llm.bytespike.ai",
+ api_key="sk-byts-...",
+ )
```
## Model id mapping
ByteSpike uses Anthropic's own model ids verbatim, plus ids for
non-Anthropic providers reachable via the Messages shape:
| Anthropic id you were using | ByteSpike id (drop-in) |
| ---------------------------- | -------------------------------------- |
| `claude-3-5-sonnet-20241022` | `claude-sonnet-4-6` (current flagship) |
| `claude-3-7-sonnet-20250219` | `claude-sonnet-4-6` |
| `claude-sonnet-4-20250514` | `claude-sonnet-4-6` |
| `claude-opus-4-20250514` | `claude-opus-4-8` |
| `claude-3-5-haiku-20241022` | `claude-haiku-4-5` |
Plus models you couldn't reach from Anthropic direct:
| Cross-vendor (Messages-API shape via translation) |
| ------------------------------------------------------------------------------- |
| `deepseek-v4-flash` |
| `deepseek-v4-pro` (translated) |
| `gemini-3-5-flash` (translated — `:translated` suffix in `protocols_aggregate`) |
| `gpt-5-5` (translated — caveats on tool schema fidelity) |
For translated models, the gateway adapts the Messages request to
the model's native protocol and the response back. See
[Endpoint types](/concepts/endpoints).
## What you gain
| Anthropic direct | ByteSpike |
| -------------------------- | ---------------------------------------------------- |
| Only Claude family | Claude + cross-vendor models via Messages shape |
| Anthropic billing | One ByteSpike wallet covers everything |
| Tier rate limits | Per-key rate limits (5h / 1d / 7d, all configurable) |
| Failures occasionally bill | Failures never bill |
| Prompt caching native | Prompt caching preserved end-to-end |
## What stays the same
* **Anthropic SDK** — Python, TypeScript, every official client
* **Messages shape** — `messages`, `system`, `tools`, `tool_choice`, `max_tokens`, `stream`, identical
* **Tool use** — `input_schema` JSON Schema format, `tool_use` blocks, `tool_result` blocks — identical
* **Prompt caching** — `cache_control` blocks on `system` / `tools` / messages — preserved end-to-end
* **Extended thinking** — `thinking` blocks on Opus / Sonnet 4.x — preserved
* **Streaming** — SSE with Anthropic event names (`message_start` / `content_block_delta` / etc) — byte-for-byte compatible
## Concrete examples
### Messages
```python theme={null}
# Was:
from anthropic import Anthropic
client = Anthropic() # reads ANTHROPIC_API_KEY
r = client.messages.create(
model="claude-3-5-sonnet-20241022",
max_tokens=1024,
messages=[{"role": "user", "content": "Hello"}],
)
# Now: env vars switched
# (ANTHROPIC_BASE_URL=https://llm.bytespike.ai, ANTHROPIC_API_KEY=sk-byts-...)
client = Anthropic()
r = client.messages.create(
model="claude-sonnet-4-6", # newer Anthropic id
max_tokens=1024,
messages=[{"role": "user", "content": "Hello"}],
)
```
### Tool use
```python theme={null}
tools = [{
"name": "get_weather",
"description": "Get current weather",
"input_schema": {
"type": "object",
"properties": {"city": {"type": "string"}},
"required": ["city"]
}
}]
r = client.messages.create(
model="claude-sonnet-4-6",
max_tokens=1024,
messages=[{"role": "user", "content": "weather in Tokyo?"}],
tools=tools,
)
```
Works identically against `claude-*`, `deepseek-v4-flash`, and
the translated routes (`gemini-3-5-flash`, `gpt-5-5` via Messages shape).
### Prompt caching
```python theme={null}
r = client.messages.create(
model="claude-sonnet-4-6",
max_tokens=1024,
system=[
{"type": "text", "text": "You are a helpful assistant."},
{
"type": "text",
"text": LARGE_KNOWLEDGE_BASE,
"cache_control": {"type": "ephemeral"},
},
],
messages=[{"role": "user", "content": "..."}],
)
# usage.cache_read_input_tokens and usage.cache_creation_input_tokens
# are populated and billed per the documented discount.
```
### Extended thinking (Opus / Sonnet 4.x)
```python theme={null}
r = client.messages.create(
model="claude-opus-4-8",
max_tokens=4096,
thinking={"type": "enabled", "budget_tokens": 2000},
messages=[{"role": "user", "content": "..."}],
)
# Response contains thinking blocks followed by text blocks.
```
## Things to double-check
Pick a routing group on the key that includes the models you
want. `claude-default` covers the Claude family. For
cross-vendor Messages-shape access (DeepSeek-Anthropic, Gemini
via translation), pick a group that includes those — usually
a multi-vendor group or the `default` group on org-tier
accounts.
Anthropic's tokenizer applies to Claude models — counts match
Anthropic direct. Translated routes (Gemini, GPT via Messages
shape) bill at the underlying model's token rate but the count
method follows that model — there can be small differences.
ByteSpike doesn't replicate Anthropic Workbench (Console UI for
prompts). If you rely on it for prompt development, develop
direct on Anthropic and deploy the prompt against ByteSpike.
Forwarded verbatim to the model. Beta features Anthropic
gates with this header work the same way through ByteSpike.
Anthropic's `/v1/messages/batches` is **not** currently exposed
on ByteSpike. Use the synchronous endpoint or our async
`/v1/tasks/*` flow for queueable work.
## Step-by-step
1. **Sign up** at [console.bytespike.ai](https://console.bytespike.ai) — see [Register](/register).
2. **Top up** \$5+ — see [Top up](/top-up).
3. **Create a key** in `claude-default` (or a multi-vendor group).
4. **Set env vars** — `ANTHROPIC_BASE_URL` + `ANTHROPIC_API_KEY` — system-wide or in `.envrc`.
5. **Run your existing script unchanged** to confirm Claude calls still work.
6. **Try cross-vendor ids** (`deepseek-v4-flash`, `gemini-3-5-flash`) one at a time.
7. **Verify caching is preserved** — `usage.cache_read_input_tokens` should populate just like Anthropic direct.
## Reverse migration
Two env-var deletes, you're back on Anthropic direct. Keep both
configs in your secrets manager if you want A/B routing.
## Next
Same idea, OpenAI side.
CLI-specific config.
Full Messages-API protocol.
How cross-protocol translation works under the hood.