TamgaEditors
An editor assistant passes through Tamga only if it can be told to send its requests to Tamga's address instead of the vendor's. Some can, some can for chat but not for inline completion, and some cannot.
No. 001Summary
"Seen" means the text reaches Tamga's scanners and the verdict applies.
| Client | Can point at Tamga | Seen | Not seen |
|---|---|---|---|
| Claude Code | Yes | Every model request: prompts, file contents, tool results, tool arguments | Its own telemetry and update checks; traffic to MCP servers |
| Continue | Yes | Chat, edit and autocomplete, per model whose apiBase is Tamga | A model configured without apiBase; embeddings and rerank models not pointed at Tamga |
| Cline | Yes | Chat and agent requests through its OpenAI Compatible or Anthropic provider | No documented header field, so no Tamga key (not verified) |
| Roo Code | Yes | Chat and agent requests through its OpenAI Compatible provider | Custom headers and the Anthropic base URL: not verified |
| Aider | Yes | Every request, over the OpenAI-compatible interface | Tamga key header: not verified |
| GitHub Copilot | Partly | Chat and agent requests to a model added as a Custom Endpoint in VS Code | Inline suggestions, which go to GitHub; semantic search and embeddings |
| Cursor | Partly | Chat with your own key and an overridden OpenAI base URL, if Tamga is reachable from the internet | Tab completion. Cursor's servers see the request before Tamga does |
| JetBrains AI Assistant | Partly | AI Chat with an OpenAI-compatible provider; inline completion if its own section points at Tamga | Features left on JetBrains' models |
| Zed | Partly | Agent and inline assistant through an OpenAI-compatible provider; edit prediction through open_ai_compatible_api | Zed's default edit prediction, and Copilot as a provider |
| Windsurf | Not verified | Nothing is known | The documentation read describes no custom endpoint. Treat as no until it does |
No. 002Two base URLs
base URL https://tamga.internal/v1
the client adds /chat/completions, /completions, /responsesbase URL https://tamga.internal/anthropic
the client adds /v1/messages, /v1/messages/count_tokensNo. 003Keys
models:
- name: Chat through Tamga
provider: openai
model: gpt-6-sol
apiBase: https://tamga.internal/v1
apiKey: <provider key>
roles: [chat, edit, apply]
requestOptions:
headers:
X-Tamga-Key: tk_your_key_hereNo. 004The rest
The browser extension does not help in an editor. A network-level check at the company's web proxy is planned and not built. What works today is a policy decision outside Tamga: allow only the clients and settings that go through it, and block the others' vendor endpoints at the firewall.