Use Claude Code with GLM, DeepSeek or Kimi#

ThinkWatch Lite connects Claude Code to a gateway on the same computer, which forwards each request to an upstream that serves the model asked for. A GLM, DeepSeek or Kimi model is either named in Claude Code, with /model or the model variables in settings.json, or reached by a routing rule that rewrites Claude’s model names to it. All three providers offer an Anthropic-compatible address, so Claude Code’s requests reach them without format conversion.

Before you start#

ThinkWatch Lite, installed, and Claude Code, run at least once.
An API key from the provider. For GLM, a Z.ai or BigModel account with a GLM Coding Plan can sign in from the app instead.
A Claude Pro or Max sign-in cannot serve as an upstream; once connected, Claude Code uses its gateway key instead.

Steps#

On the Upstreams page, choose New upstream. The first step, Service, lists the services in three groups, with a search box; picking one goes straight to Connection, which asks only for what that service needs:
- DeepSeek fills in https://api.deepseek.com/anthropic and the protocol. Enter the API key.
- GLM with a key: Z.ai / BigModel. Choose the Site, BigModel (mainland China) or Z.ai (international), which sets the Anthropic-compatible address, and enter the API key. GLM’s OpenAI-compatible address also works from ThinkWatch Lite 2026.10.5, with Custom and Protocol set to OpenAI Chat Completions: https://api.z.ai/api/paas/v4 or https://open.bigmodel.cn/api/paas/v4, and …/api/coding/paas/v4 on a GLM Coding Plan. Claude Code’s requests are then converted, so the Anthropic address is the more direct choice.
- GLM with an account: Z.ai / BigModel, with Authentication set to Account sign-in. Choose the Site, tick Acknowledge the notes above, choose Sign in with the browser and authorize; to sign in from a browser other than the default one, choose Copy link instead. The app creates an API key named thinkwatch on the account and saves the upstream.
- Kimi: Custom, with the Anthropic-compatible base URL from Kimi’s documentation and Protocol set to Anthropic Messages. For Kimi For Coding, also turn on Forward client identity under Advanced settings.
With a key, Next verifies the address and key and fetches the model list at no cost; choose Next twice more, then Create. After an account sign-in the upstream is already saved and the dialog is on Models; choose Next, then Done.
On the Clients page, choose Connect… on the Claude Code row. The dialog lists the fields that change in ~/.claude/settings.json: env.ANTHROPIC_BASE_URL, env.ANTHROPIC_AUTH_TOKEN (a new key named claude-code) and env.CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY. Choose Connect.
Choose how the model is selected:
- By name. In Claude Code, /model <model ID> switches to the model, and claude --model <model ID> starts with it. To map Claude Code’s model aliases to it, add these to the env block of ~/.claude/settings.json; the Haiku one also runs background tasks.
```
"ANTHROPIC_DEFAULT_OPUS_MODEL": "<model ID>",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "<model ID>",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "<model ID>"
```
  No routing rule is needed; the default route skips upstreams whose model list does not contain it.
- By rewriting. On the Routing page, open the default route and choose Add rule. With Add condition, add Model claude-*, and Key claude-code to leave other clients alone. Set On match to Forward, Forward to to the upstream, and Change model to under Parameter rewrites to the model ID. Choose Add, then Save. Claude Code keeps showing Claude’s names.

Notes#

Format conversion. Claude Code sends Anthropic Messages; with that protocol the request goes out unchanged, apart from identity fields such as metadata.user_id, which the gateway removes. Only an upstream in another format, such as an OpenAI-compatible address set to OpenAI Chat Completions, gets a converted request: Traffic marks it Converted and its details list any dropped fields. Web search, a server-side tool, cannot be converted and is not sent to such an upstream. Auto-detect does not recognize these addresses and forwards requests in the client’s own format, which suits Claude Code but not Codex.
Forward client identity is off by default, so upstreams see ThinkWatch’s User-Agent and no client identity. Kimi For Coding, Bailian Coding Plan and similar upstreams accept only certain clients; with the switch on, they receive Claude Code’s own User-Agent, identity headers such as x-app and the identity fields in the body, unaltered.
GLM Coding Plan. An upstream on api.z.ai or open.bigmodel.cn, signed in or added with a key, shows its 5-hour and weekly limits, and the credits left on a plan billed in credits, in the Quota / billing column and in the menu bar or tray menu.
Balance. A DeepSeek upstream, and a Kimi upstream on api.moonshot.cn or api.moonshot.ai, shows the account balance in the Quota / billing column.
The /model list shows gateway models only when their names contain claude or anthropic. Other models are typed by name, or added as one entry with ANTHROPIC_CUSTOM_MODEL_OPTION.
Cost. A rewritten request is priced as the model actually sent.
Restore… puts back only the fields the app wrote; model variables added by hand stay in settings.json.

Last updated 2026-10-10 Edit on GitHub