Skip to main content
Connect PyCharm AI Assistant to Token Factory through its native OpenAI-compatible provider. You can use it to read project files, suggest edits, and continue working in the same conversation. PyCharm 2026.2.2 with AI Assistant 262.10315.187 passed five read/edit turns with zai-org/GLM-5.3 and the gateway tool-argument compatibility fix. This validation used a local patched gateway connected to our development endpoint; it does not establish availability of the fix on every endpoint. For a separate agent route, see Claude Code and ACP.

​
Configure the native provider

You need PyCharm with the JetBrains AI Assistant plugin, a Token Factory API key, and a model ID from the model catalog. The settings below follow the reported PyCharm 2026.2.2 configuration; labels can differ between IDE versions.
  1. Open Settings → Tools → AI Assistant → Third-party AI providers.
  2. Add an OpenAI-compatible provider.
  3. Set the URL to https://api.tokenfactory.corvex.cloud/v1 and enter your Token Factory API key.
  4. Select zai-org/GLM-5.3 for Core features. When checking tool calling, use the same model for Instant helpers so both request types use the intended provider.
  5. Enable Tool calling when evaluating file reads or edits.
Start with a disposable project. Ask AI Assistant to read a file and make a small edit, then continue the same conversation with another read and edit. In native Chat, use Apply, Accept, or Create File when offered, then save and inspect the files. A suggested snippet is not a saved file change.

​
Follow-up tool calls return 400

The affected client can encode function.arguments twice when it replays an assistant tool call. The next request may fail with:
This issue is tracked by JetBrains as LLM-30278. An upstream code fix alone does not confirm that a particular installed plugin contains it. The gateway compatibility fix removes one redundant encoding layer when the result is a valid JSON object. If your endpoint still returns this error, contact support with your PyCharm build, AI Assistant plugin version, model ID, error text and request ID if available so we can check your endpoint’s status. Do not include your API key or private project contents.

​
Concurrent requests return 429

AI Assistant can send several helper requests alongside the main chat request. A 429 Concurrent request limit exceeded error is separate from the tool argument failure above. Respect Retry-After, avoid immediately repeating the whole request burst. This message can come from your key’s concurrency limit or a model capacity guard; it does not by itself mean your key needs a higher limit. Ask support to review the error text, timing and request ID so we can identify the guard involved. Other 429 errors can have different quota causes.

​
Claude Code and ACP

The Claude Code guide provides an isolated claude-corvex launcher for Token Factory. Verify that configuration in a terminal first. PyCharm can also run external agents through the Agent Client Protocol (ACP). Connecting Claude Code this way requires an ACP-compatible adapter and its configuration; the terminal launcher alone is not an ACP server. Follow JetBrains’ custom-agent setup and the adapter’s instructions, and confirm it uses your Token Factory endpoint and model. Contact support for the Token Factory configuration appropriate to your IDE and adapter versions. This is a separate path from AI Assistant’s native OpenAI-compatible provider.
Last modified on October 8, 2026