Free-ZAI-Api

Self-hosted OpenAI-compatible API proxy for chat.z.ai (GLM-5.3 / GLM-5.2 / etc). Runs a real browser session via Playwright, so the site sees a genuine Chrome fingerprint — the invisible Aliyun captcha passes automatically, no third-party captcha services needed.

Built for use with opencode and any other OpenAI-compatible client.

client (opencode)                proxy (main.py)                  chat.z.ai
      │  POST /v1/chat/completions     │                               │
      │  {tools, messages, ...}        │                               │
      ├───────────────────────────────>│  Playwright browser session   │
      │                                ├──────────────────────────────>│
      │                                │   fresh chat + model select   │
      │                                │   invisible captcha auto-pass │
      │                                │<──────────────────────────────┤
      │<───────────────────────────────┤  SSE stream                   │
      │  OpenAI chunks                 │  (reasoning_content + content)│

• Tool calling — z.ai has no native function-calling over web chat, so the proxy injects tool schemas into the prompt and streams back the model's <tc>NAME<ak>key</ak><av>value</av></tc> responses as standard OpenAI delta.tool_calls (z.ai strips native <tool_call>/<arg_key>/<arg_value> tags from the stream, hence the custom tc/ak/av tags).
• Account rotation — round-robin between accounts from accounts.json.
• Cooldown — configurable delay between requests to avoid triggering captchas.
• Popups — promotional dialogs are closed automatically by a background watcher.
• Startup screen — plain printed output: banner, option table, the Choice: line, then the usage counters, and a single CSI n A afterwards to walk the caret back up to the prompt. The rows the move crosses are counted from the block that was just printed, so there is no cursor arithmetic that can drift. Two details are load-bearing:
  • a terminal narrower than the banner is asked to grow via XTWINOPS, and the wait is a poll on the actual width rather than a fixed pause. A fixed pause lost that race: the first frame was laid out against the old width and the banner came out clipped. The request is for the banner width plus a margin, so the logo is not flush against both edges either;
  • a counter row is printed one column short of the terminal width. A row that fills the width exactly leaves the cursor in pending-wrap, and the row count the caret move depends on stops being reliable.
• Usage counters — tokenised with deepseek-tokenizer, real token counts (prompt, completion, reasoning) next to character counts, split into GLM Chat and GLM Reasoning. They accumulate in stats.json next to the script, are reloaded on every start, and each row's bar shows that group's share of the same counter across everything, so All is always 100%. Tokenising happens on a background thread: a large prompt costs about a second, and the request should not wait for it. An empty or missing stats.json draws no block at all rather than a row of zeroes with full bars.
• Scrolling the counters — if they do not fit the window they are cut to it and the heading says so, as Stats (scroll):. The mouse wheel or up/down and PageUp/PageDown move the block one row at a time and the heading stays pinned. On Windows the wheel comes from the console as a mouse event rather than as characters, so mouse reporting is only switched on while there is something to scroll - that is what turns Quick Edit off, and with it off the terminal cannot select text; hold Shift while dragging to select anyway. A wheel step repaints the counter band in place and never clears the screen, otherwise reprinting the banner - a colour code per character - leaves a visible blank frame.

git clone https://github.com/lothiann/Free-ZAI-Api.git
cd Free-ZAI-Api
pip install -r requirements.txt
playwright install chromium

Get your auth token: open https://chat.z.ai → DevTools (F12) → Console:

fetch('/api/v1/auths/').then(r=>r.json()).then(d=>window.__acc={id:d.id,email:d.email,name:d.name,token:d.token})

copy(JSON.stringify(__acc))

Paste into accounts.json:

{
  "rotate_every": 10,
  "accounts": [
    { "name": "main", "email": "you@gmail.com", "token": "eyJ..." }
  ]
}

python main.py

Server starts on http://127.0.0.1:8492/v1.

curl http://127.0.0.1:8492/v1/models

curl -N http://127.0.0.1:8492/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d "{\"model\":\"glm-5.2\",\"stream\":true,\"messages\":[{\"role\":\"user\",\"content\":\"say OK\"}]}"

# account status
curl http://127.0.0.1:8492/accounts

Add to ~/.config/opencode/opencode.jsonc (or opencode.json):

{
  "provider": {
    "zai": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Z.AI (browser proxy)",
      "options": {
        "baseURL": "http://127.0.0.1:8492/v1",
        "apiKey": "sk-nothing"
      },
      "models": {
        "glm-5.3": { "name": "GLM-5.3", "limit": { "context": 2000000, "output": 2000000 } },
        "x-preview-l": { "name": "GLM-5.3 Flash", "limit": { "context": 2000000, "output": 2000000 } },
        "glm-5.2": { "name": "GLM-5.2", "limit": { "context": 2000000, "output": 2000000 } },
        "GLM-5-Turbo": { "name": "GLM-5 Turbo", "limit": { "context": 2000000, "output": 2000000 } },
        "GLM-5v-Turbo": { "name": "GLM-5v Turbo", "limit": { "context": 2000000, "output": 2000000 } },
        "glm-4.7": { "name": "GLM-4.7", "limit": { "context": 2000000, "output": 2000000 } },
        "glm-4.6v": { "name": "GLM-4.6v", "limit": { "context": 2000000, "output": 2000000 } },
        "GLM-4.1V-Thinking-FlashX": { "name": "GLM-4.1V Thinking FlashX", "limit": { "context": 2000000, "output": 2000000 } },
        "deep-research": { "name": "Deep Research", "limit": { "context": 2000000, "output": 2000000 } },
        "zero": { "name": "Zero", "limit": { "context": 2000000, "output": 2000000 } },
        "0727-106B-API": { "name": "0727 106B API", "limit": { "context": 2000000, "output": 2000000 } },
        "0727-360B-API": { "name": "0727 360B API", "limit": { "context": 2000000, "output": 2000000 } },
        "0808-360B-DR": { "name": "0808 360B DR", "limit": { "context": 2000000, "output": 2000000 } },
        "glm-4-flash": { "name": "GLM-4 Flash", "limit": { "context": 2000000, "output": 2000000 } },
        "glm-4-air-250414": { "name": "GLM-4 Air", "limit": { "context": 2000000, "output": 2000000 } }
      }
    }
  }
}

Models can also be discovered automatically from /v1/models, but listing them manually is more reliable. Passing reasoning: true on a model enables default/high/max thinking-effort variants (mapped to the site's Deep Think control).

• Each request opens a fresh site chat; full conversation history is forwarded as a single prompt.
• Reasoning is streamed via delta.reasoning_content, the answer via delta.content.
• Tool calls require the client to send standard OpenAI tools; results come back as role: "tool" messages.
• usage counts characters (1 token := 1 char), matching the site's ~2M real limit — set the model limit in your client accordingly (e.g. context: 2000000).
• Logs are written to logs/, the last raw response to last_response.json.

For personal / educational use. Automating chat.z.ai may violate its Terms of Service — use your own accounts at your own risk.