# BifrostTransform

**Özet:** Anthropic Messages API formatı ile OpenAI Chat Completions formatı arasında çift yönlü dönüşüm yapan katman (`internal/transform/anthropic_bifrost.go:74`). "Bifrost" (İskandinav mitolojisindeki köprü) adı, two-way dönüşümü simgeler.

**Kütüphaneler:** Go `encoding/json`, `strings`

**Bağlantılar:** [[AnthropicHandler]], [[Index]]

## Struct'lar

### Anthropic Format (Giriş/Çıkış)

| Struct | Alanlar |
|---|---|
| `AnthropicRequest` | `Model`, `MaxTokens`, `Messages []AnthropicMessage`, `System`, `Stream` |
| `AnthropicMessage` | `Role`, `Content` |
| `AnthropicResponse` | `ID`, `Type`, `Role`, `Content []AnthropicContent`, `Model`, `StopReason`, `StopSequence` |
| `AnthropicContent` | `Type`, `Text` |

### OpenAI / Bifrost Format (Ara Katman)

| Struct | Alanlar |
|---|---|
| `BifrostRequest` | `Model`, `Messages []BifrostMessage`, `MaxTokens`, `Stream` |
| `BifrostMessage` | `Role`, `Content` |
| `BifrostResponse` | `ID`, `Object`, `Created`, `Model`, `Choices[{Message{Role,Content}, FinishReason, Index}]` |

## Dönüşüm Detayları

### Anthropic → OpenAI (`AnthropicToBifrost()`)

```
Anthropic Request                     OpenAI Request
─────────────────                     ─────────────
model: "claude-3-5-sonnet"     →     model: "Anthropic/claude-3-5-sonnet"
system: "Be helpful"           →     messages: [{role:"system", ...}, ...]
messages: [{role, content}]    →     messages: [{role, content}]
max_tokens: 1024               →     max_tokens: 1024
stream: true                   →     stream: true
```

### OpenAI → Anthropic (`BifrostToAnthropic()`)

```
OpenAI Response                       Anthropic Response
───────────────                       ─────────────────
id: "chatcmpl-xxx"             →     id: "chatcmpl-xxx"
choices[].message.content      →     content: [{type:"text", text:"..."}]
choices[].finish_reason        →     stop_reason: "stop"/"end_turn"
model                          →     model (same)
```

## Model Prefix Tahmini (`guessProvider()`)

| Model içerir | Prefix | Örnek |
|---|---|---|
| `deepseek` | `DeepSeek/` | `deepseek-v4` → `DeepSeek/deepseek-v4` |
| `gpt`, `openai` | `OpenAI/` | `gpt-4` → `OpenAI/gpt-4` |
| `claude`, `anthropic` | `Anthropic/` | `claude-3` → `Anthropic/claude-3` |
| `gemini` | `Google/` | `gemini-pro` → `Google/gemini-pro` |
| diğer | `DeepSeek/` (varsayılan) | — |

## SSE Streaming Transformer (`StreamTransformer`)

OpenAI SSE chunk'larını Anthropic SSE event'lerine dönüştürür (`TransformChunk()`):

```
OpenAI SSE Chunk                    Anthropic SSE Events
─────────────────                    ─────────────────────
id + model var →                    message_start (ilk chunk'ta)
delta.content →                     content_block_start + content_block_delta
finish_reason →                     content_block_stop + message_delta + message_stop
[DONE] →                            (finish() zaten çağrıldıysa no-op)
```

**Buffer:** 4MB scanner buffer (`handlers/anthropic.go`) — Claude Code tool çağrılarında büyük chunk'lar için.
**Hata yönetimi:** Scanner/write/flush hataları `[ANTHROPIC]` prefix'iyle loglanır.