https://github.com/NIyueeE/ds-free-api

DeepSeek网页端API代理，支持OpenAI与Anthropic兼容接口 | OpenAI & Anthropic compatible API proxy for DeepSeek web
https://github.com/NIyueeE/ds-free-api
api api-proxy claude-api deepseek-api freeapi openai-api proxy-server rust
Last synced: 2 days ago
JSON representation
DeepSeek网页端API代理，支持OpenAI与Anthropic兼容接口 | OpenAI & Anthropic compatible API proxy for DeepSeek web
Host: GitHub
URL: https://github.com/NIyueeE/ds-free-api
Owner: NIyueeE
License: gpl-3.0
Created: 2026-03-21T15:17:34.000Z (2 months ago)
Default Branch: main
Last Pushed: 2026-05-26T16:27:26.000Z (6 days ago)
Last Synced: 2026-05-26T18:19:12.877Z (6 days ago)
Topics: api, api-proxy, claude-api, deepseek-api, freeapi, openai-api, proxy-server, rust
Language: Rust
Homepage:
Size: 2.03 MB
Stars: 356
Watchers: 6
Forks: 101
Open Issues: 15
Metadata Files:
- Readme: README.en.md
- Changelog: CHANGELOG.md
- License: LICENSE
- Codeowners: .github/CODEOWNERS
- Agents: AGENTS.md
Awesome Lists containing this project

README

          


  



DS-Free-API




  

  

  

  





  

  

  



[中文](README.md)

A Rust API proxy that translates DeepSeek's free web chat into standard OpenAI and Anthropic-compatible API protocols (supports chat completions and messages, including streaming and tool calling).

## Highlights

- **Zero-cost API proxy**: Uses DeepSeek's free web interface — no official API key needed, get OpenAI/Anthropic-compatible endpoints for free

- **Dual protocol support**: Both OpenAI Chat Completions and Anthropic Messages API, drop-in compatible with mainstream clients

- **Tool call ready**: Full OpenAI function calling implementation with a 3-tier self-healing pipeline (text repair → JSON repair → model fallback), covering 10+ malformed formats

- **File upload ready**: Inline data URL files in OpenAI `file`/`image_url` content parts and Anthropic `image`/`document` content blocks are automatically uploaded to DeepSeek sessions; HTTP URLs trigger search mode so the model can access link content directly

- **Oversized prompt fallback**: When the prompt exceeds model limits, automatically falls back to chunked completion with file upload

- **Web admin panel**: Built-in dashboard for account pool status, API key management, request logs, i18n internationalization, theme switcher, and hot-reloadable config — ready out of the box

- **Built with Rust**: Single binary + single TOML config, cross-platform native performance (web panel compiled in at build time)

- **Multi-account pool**: Idle-aware round-robin selection (DashMap lock-free reads), horizontal scaling for concurrency

## Quick Start

### Binary Usage

1. Download and extract the archive for your platform from [releases](https://github.com/NIyueeE/ds-free-api/releases)

2. Copy `config.example.toml` to `config.toml` and fill in accounts (optional — you can also configure via the admin panel after startup)

3. Run `./ds-free-api`

4. Visit `http://127.0.0.1:22217/admin` to set an admin password, then manage API keys and accounts from the panel

```bash

./ds-free-api

./ds-free-api -c /path/to/config.toml

RUST_LOG=debug ./ds-free-api

```

> **Concurrency**: The free API has session-level rate limits. This project has built-in rate-limit detection + exponential backoff retry for stability.

> Recommended parallelism = accounts / 2. Supports starting without `config.toml` and adding accounts via the admin panel.

### Docker Usage

```bash

docker compose -f docker/docker-compose.yaml up -d

```

Refer to the [sample compose file](./docker/docker-compose.yaml) for reference.

The admin panel is at `http://localhost:22217/admin`. Set your admin password on first visit.

The `config/` and `data/` directories are bind-mounted into the container — config changes persist to the host automatically.

### Free Test Accounts

Please register your own. You can refer to the method in [issue #62](https://github.com/NIyueeE/ds-free-api/issues/62).

## API Endpoints

| Method | Path | Description |

|--------|------|-------------|

| GET    | `/`   | Redirect to admin panel |

| GET    | `/health` | Health check |

| POST   | `/v1/chat/completions` | Chat completions (streaming + tool calls) |

| GET    | `/v1/models` | List models |

| GET    | `/v1/models/{id}` | Model details |

| POST   | `/anthropic/v1/messages` | Anthropic Messages (streaming + tool calls) |

| GET    | `/anthropic/v1/models` | List models (Anthropic format) |

| GET    | `/anthropic/v1/models/{id}` | Model details (Anthropic format) |

The admin panel is at `/admin` — on first visit you'll be guided to set an admin password.

## Model Mapping

The `model_types` config in `config.toml` (default `["default", "expert", "vision"]`) maps to model IDs:

| OpenAI Model ID    | DeepSeek Mode  |

| ------------------ | -------------- |

| `deepseek-default` | Fast mode      |

| `deepseek-expert`  | Expert mode    |

| `deepseek-vision`  | Vision mode    |

Optional aliases via `model_aliases`, aligned by index with `model_types`. Empty strings are skipped:

```toml

# model_aliases = ["", "deepseek-v4-pro"]  → deepseek-v4-pro maps to expert (index 1)

model_aliases = []

```

The Anthropic compatibility layer uses the same model IDs via `/anthropic/v1/messages`.

### Capability Toggles

- **Deep thinking**: Enabled by default. To explicitly disable, include `"reasoning_effort": "none"` in the request body.

- **Web search**: Enabled by default (DeepSeek injects a stronger system prompt in search mode, improving tool call adherence). To explicitly disable, include `"web_search_options": {"search_context_size": "none"}` in the request body.

- **File upload**: Inline files (data URL) are auto-uploaded to DeepSeek sessions; HTTP URLs trigger search mode:

  **OpenAI endpoint:**

  ```json

  {"type": "file", "file": {"file_data": "data:text/plain;base64,...", "filename": "doc.txt"}}

  {"type": "image_url", "image_url": {"url": "data:image/png;base64,..."}}

  {"type": "image_url", "image_url": {"url": "https://example.com/img.jpg"}}

  ```

  **Anthropic endpoint:**

  ```json

  {"type": "image", "source": {"type": "base64", "media_type": "image/png", "data": "..."}}

  {"type": "document", "source": {"type": "base64", "media_type": "text/plain", "data": "..."}}

  {"type": "image", "source": {"type": "url", "url": "https://example.com/img.jpg"}}

  ```

### Tool Call Tag Hallucination

Built-in fuzzy matching handles variations (full-width `｜`<=>`|`, `▁`<=>`_`) for most formats. If the model outputs a different fallback tag, add it via the admin panel or in `config.toml` under `[ds_core]`:

```toml

tool_call.extra_starts = ["<|tool_call_begin|>", "", ""]

tool_call.extra_ends = ["<|tool_call_end|>", "", ""]

```

## Web Admin Panel

Visit `http://127.0.0.1:22217/admin` after starting the server:

- **Dashboard**: Request statistics, account pool status at a glance

- **Accounts**: View/add/remove accounts, manually re-login accounts in Error state

- **API Keys**: Create/delete API keys, masked display

- **Models**: Available models with details

- **Config**: Current runtime config (sensitive fields masked)

- **Logs**: Recent request logs and runtime logs



  

  


  Dashboard overview





  

  


  Config editor page



On first visit, you'll be guided to set an admin password (stored as bcrypt hash), then issued a JWT (24h validity). Password reset revokes old tokens.

## Environment Variables

| Variable | Default | Description |

|----------|---------|-------------|

| `RUST_LOG` | `info` | Log level (`trace` / `debug` / `info` / `warn` / `error`) |

| `DS_DATA_DIR` | `.` (current dir) | Data directory for `logs/runtime.log` and `stats.json` |

| `DS_CONFIG_PATH` | `./config.toml` | Config file path (lower priority than `-c` flag) |

## Security

- **Admin panel**: JWT authentication + bcrypt password hash + login rate limiting (5 failures → 5-minute lockout)

- **API access**: API keys created via the admin panel (HashSet O(1) lookup)

- **CORS**: Configurable allowed origins, defaults to `http://localhost:22217`

- **Sensitive data**: Account IDs masked in response headers, request bodies excluded from logs, persisted files at 0600 permissions

## Development

### Design Philosophy

**A single `config.toml` reflects all runtime state.** Admin panel changes are instantly persisted to `config.toml` and hot-reloaded into the running service.

**No unnecessary runtime system dependencies.** The project prioritizes pure Rust or statically-linked dependencies (e.g., `rustls` → `wreq` with BoringSSL), ensuring a single binary with no external `.so`/`.dll` requirements — download and run.

### Architecture Diagram

```mermaid

flowchart TB

    %% ===== Theme =====

    classDef client fill:#eff6ff,stroke:#3b82f6,stroke-width:3px,color:#1d4ed8,rx:14,ry:14

    classDef gateway fill:#fffbeb,stroke:#f59e0b,stroke-width:3px,color:#92400e,rx:12,ry:12

    classDef openai_adapter fill:#f8fafc,stroke:#0a9e7b,stroke-width:2px,color:#334155,rx:10,ry:10

    classDef anthropic_compat fill:#f8fafc,stroke:#d07354,stroke-width:2px,color:#334155,rx:10,ry:10

    classDef ds_core fill:#f8fafc,stroke:#3964fe,stroke-width:2px,color:#1e40af,rx:10,ry:10

    classDef external fill:#fef2f2,stroke:#ef4444,stroke-width:3px,color:#991b1b,rx:6,ry:6

    %% ===== Nodes =====

    Client(["Client"]):::client

    subgraph GW ["HTTP Gateway Layer"]

        Handler(["Router / Auth / Serialization"]):::gateway

    end

    subgraph PL ["Protocol Layer"]

        direction TB

        subgraph AC ["Anthropic Compat"]

            A2O["Request
Anthropic → OpenAI"]:::anthropic_compat

            O2A["Response
OpenAI → Anthropic"]:::anthropic_compat

        end

        subgraph OA ["OpenAI Adapter"]

            ReqPipe["Request Pipeline
Validation / Tool Extraction / Prompt Building"]:::openai_adapter

            RespPipe["Response Pipeline
SSE Parsing / Format Conversion / Tool Repair"]:::openai_adapter

        end

    end

    subgraph CL ["Core Layer (ds_core)"]

        Pool["Account Pool Rotation"]:::ds_core

        PoW["PoW Solver"]:::ds_core

        Session["Session Orchestration
Create/Destroy / History Upload"]:::ds_core

    end

    DeepSeek[("DeepSeek API")]:::external

    %% ===== Connections =====

    Client -->|"HTTP Request"| Handler

    Handler -->|"OpenAI Request"| ReqPipe

    Handler -->|"Anthropic Request"| A2O

    A2O -->|"OpenAI Request"| ReqPipe

    ReqPipe --> Pool

    Pool --> PoW

    PoW --> Session

    Session -->|"completion endpoint"| DeepSeek

    Session -.->|"DeepSeek SSE Stream"| RespPipe

    RespPipe -.->|"OpenAI Response"| Handler

    RespPipe -.->|"OpenAI Response"| O2A

    O2A -.->|"Anthropic Response"| Handler

    %% ===== Subgraph Styles =====

    style GW fill:#fffbeb,stroke:#f59e0b,stroke-width:2px,stroke-dasharray: 5 5

    style PL fill:#fafafa,stroke:#94a3b8,stroke-width:2px

    style AC fill:#fdf0ec,stroke:#d07354,stroke-width:2px

    style OA fill:#e6f7f3,stroke:#0a9e7b,stroke-width:2px

    style CL fill:#eef2ff,stroke:#3964fe,stroke-width:2px,stroke-dasharray: 5 5

```

### Data Pipeline

#### OpenAI (chat_completions) Pipeline:

```mermaid

flowchart TB

    %% ===== Theme =====

    classDef ds_core fill:#eef2ff,stroke:#3964fe,stroke-width:2.5px,color:#1e40af,rx:10,ry:10

    classDef openai_adapter fill:#e6f7f3,stroke:#0a9e7b,stroke-width:2.5px,color:#065f46,rx:10,ry:10

    classDef step fill:#fffbeb,stroke:#f59e0b,stroke-width:1.5px,color:#334155,rx:6,ry:6

    subgraph RQ ["Request Pipeline"]

        direction TB

        Q1["ChatCompletionsRequest"]:::openai_adapter

        Q2["Validation + Defaults"]:::step

        Q3["Extract tools/files + inject prompts"]:::step

        Q4["Build DeepSeek native tag prompt"]:::step

        Q5["Model mapping + capability toggles"]:::step

        Q6["Retry with exp. backoff
1s→2s→4s→8s→16s"]:::step

        Q7["ChatRequest"]:::ds_core

    end

    subgraph RS1 ["Non-streaming Response"]

        direction TB

        OR1["ds_core SSE stream"]:::ds_core

        OR2["SSE frame parse
ContentDelta / Usage"]:::step

        OR3["State machine merge
contiguous text / accumulate usage"]:::step

        OR4["Chunk aggregation
concat content / reasoning / tool_calls"]:::step

        OR5["ChatCompletionsResponse"]:::openai_adapter

    end

    subgraph RS2 ["Streaming Response"]

        direction TB

        OS1["ds_core SSE stream"]:::ds_core

        OS2["SSE frame parse + state machine"]:::step

        OS3["Chunk conversion
DsFrame → ChatCompletionsResponseChunk"]:::step

        OS4["Tool call XML parse"]:::step

        OS5["Malformed tool call repair"]:::step

        OS6["Stop sequence detect + obfuscation"]:::step

        OS7["ChatCompletionsResponseChunk"]:::openai_adapter

    end

    Q1 --> Q2 --> Q3 --> Q4 --> Q5 --> Q6 --> Q7

    OR1 --> OR2 --> OR3 --> OR4 --> OR5

    OS1 --> OS2 --> OS3 --> OS4 --> OS5 --> OS6 --> OS7

    style RQ fill:#f8fafc,stroke:#0a9e7b,stroke-width:2px

    style RS1 fill:#f8fafc,stroke:#0a9e7b,stroke-width:2px

    style RS2 fill:#f8fafc,stroke:#0a9e7b,stroke-width:2px

```

#### Anthropic (messages) Pipeline:

```mermaid

flowchart TB

    %% ===== Theme =====

    classDef oai fill:#e6f7f3,stroke:#0a9e7b,stroke-width:2.5px,color:#065f46,rx:10,ry:10

    classDef anth fill:#fdf0ec,stroke:#d07354,stroke-width:2.5px,color:#7c3a2a,rx:10,ry:10

    classDef step fill:#fffbeb,stroke:#f59e0b,stroke-width:1.5px,color:#334155,rx:6,ry:6

    subgraph RQ ["Request Pipeline"]

        direction TB

        Q1["MessagesRequest"]:::anth

        Q2["Message expansion
System prepend / text merge / image+document mapping"]:::step

        Q3["Tool mapping
ToolUnion → OpenAI Tool"]:::step

        Q4["Capability toggle mapping
thinking → reasoning_effort"]:::step

        Q5["ChatCompletionsRequest"]:::oai

    end

    subgraph RS3 ["Non-streaming Response"]

        direction TB

        AR1["ChatCompletionsResponse"]:::oai

        AR2["Content decomposition
reasoning → Thinking
content → Text
tool_calls → ToolUse"]:::step

        AR3["ID mapping
chatcmpl → msg
call → toolu"]:::step

        AR4["MessagesResponse"]:::anth

    end

    subgraph RS4 ["Streaming Response"]

        direction TB

        AS1["ChatCompletionsResponseChunk stream"]:::oai

        AS2["Chunk state machine
block type switch / index progression"]:::step

        AS3["Event mapping
content → text_delta
reasoning → thinking_delta
tool_calls → input_json_delta"]:::step

        AS4["MessagesResponseChunk"]:::anth

    end

    Q1 --> Q2 --> Q3 --> Q4 --> Q5

    AR1 --> AR2 --> AR3 --> AR4

    AS1 --> AS2 --> AS3 --> AS4

    style RQ fill:#f8fafc,stroke:#d07354,stroke-width:2px

    style RS3 fill:#f8fafc,stroke:#d07354,stroke-width:2px

    style RS4 fill:#f8fafc,stroke:#d07354,stroke-width:2px

```

For detailed development guide (building, testing, Docker deployment, e2e testing, etc.), see [docs/development.md](./docs/development.md).

## License

[GNU General Public License v3.0](LICENSE)

[DeepSeek's official API](https://platform.deepseek.com/top_up) is very affordable — please support the official service.

This project was born from the desire to try the latest models in DeepSeek's web interface during grayscale testing.

**Commercial use is strictly prohibited** to avoid putting pressure on official servers. Use at your own risk.
ecosyste.ms

Data

Tools

Indexes

Applications

Experiments

Awesome

https://github.com/NIyueeE/ds-free-api

Awesome Lists containing this project

README

DS-Free-API
