Skip to content

feat: generic background Responses API providers — sference backend #32

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
apejcic wants to merge 1 commit into
base: main
Choose a base branch
from
Open

feat: generic background Responses API providers — sference backend #32

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
45 changes: 44 additions & 1 deletion README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,6 +89,7 @@ Claude Code reads these environment variables to determine where to send API cal
| **DeepSeek** (default) | `--backend ds` | $0.44 | $0.87 | China | Auto context caching (120x cheaper on repeat turns) |
| **OpenRouter** | `--backend or` | $0.44 | $0.87 | US | Cheapest, lowest latency from US/EU |
| **Fireworks AI** | `--backend fw` | $1.74 | $3.48 | US | Fastest inference |
| **sference** | `--backend sf -w 1h` | (varies) | (varies) | EU | Async background inference, up to 75% off |
| **Anthropic** | `--backend anthropic` | $3.00 | $15.00 | US | Original Claude Opus (for hard problems) |

### Setup per backend
Expand All @@ -111,7 +112,49 @@ setx FIREWORKS_API_KEY "fw_..." # Windows
export FIREWORKS_API_KEY="fw_..." # macOS/Linux
```

## Cost comparison
**sference** (async background — lower cost, higher latency per turn):
```bash
export SFERENCE_API_KEY="sk_..." # macOS/Linux
export SFERENCE_MODEL="moonshotai/Kimi-K2.6" # optional, this is the default
export SFERENCE_COMPLETION_WINDOW="1h" # 1h | 24h (sference); others passed to custom providers
deepclaude -b sf -w 1h
```

**Custom background provider** (any Responses API with `background: true`):
```bash
export BG_PROVIDER_URL="https://api.example.com"
export BG_PROVIDER_API_KEY="sk_..."
export BG_PROVIDER_MODEL="your-model"
export BG_PROVIDER_WINDOW="1h"
deepclaude -b bg
```

## Background providers (`responses-bg`)

Some inference platforms expose an async **Responses API**: submit with `background: true`, poll until complete. Claude Code only speaks Anthropic `/v1/messages`, so deepclaude's proxy translates each turn:

```
Claude Code → POST /v1/messages (Anthropic format, streaming)
Proxy → POST /v1/responses (background: true, completion_window)
Proxy → poll GET /v1/responses/{id}
Proxy → synthesize Anthropic SSE back to Claude Code
```

**Trade-off:** each agent turn waits in the provider's queue (minutes to hours depending on window). Best for overnight headless runs (`claude -p`) or when cost matters more than latency. Sync backends (DeepSeek, OpenRouter) remain the default for interactive coding.

Control endpoints while the proxy runs:
```bash
curl -s http://127.0.0.1:3200/_proxy/status # mode, window, tool_mode
curl -sX POST http://127.0.0.1:3200/_proxy/window -d "window=24h"
curl -s http://127.0.0.1:3200/_proxy/cost
```

Add a slash command in `~/.claude/commands/sference.md`:
```
Switch to sference background mode:
curl -sX POST http://127.0.0.1:3200/_proxy/mode -d "backend=sference"
If successful, say: "Switched to sference (background)."
```

| Usage level | Anthropic Max | deepclaude (DeepSeek) | Savings |
|---|---|---|---|
Expand Down
Loading