Select a provider from the list, or add a new one.

Git Commit

Model for Git commit message generation.
Language for generated messages.

Global Settings

You don't need this — registering custom models (BYOK) already lets you use Copilot Chat while signed out of GitHub. This toggle sets VS Code's experimental chat.allowAnonymousAccess setting, which is hidden, experiment-gated, and marked for removal; it may not be available in your VS Code version. Leave it off unless you know you need it.
By default VS Code clears the open chat panel on a fresh start, so your conversations disappear after restarting (especially when signed out of GitHub). Enabling this sets chat.restoreLastPanelSession to true so the last chat session is restored from disk on startup. Chat history files are always written to disk; this only controls whether they are re-opened.
VS Code reports the selected model id in its core chat telemetry (interactiveSessionProviderInvoked), which can reveal which third-party API providers and models you use. Enabling this sets telemetry.telemetryLevel to off. Recommended for anonymity.
When a request to a provider fails, print a reproducible curl command and the server's response body to the Developer Tools console (Help > Toggle Developer Tools). Sensitive headers (API keys, tokens) are masked. Sets customcopilot.debugRequestLogging. Useful for debugging request/response problems.
This extension is a provider: Copilot Chat builds the full system prompt (its identity, safety rules, tool instructions) and sends it here. Enable this to rewrite that system prompt before it is forwarded to your endpoint — replace or extend the default prompt and its parts (e.g. stricter safety rules) for all models. Sets customcopilot.promptOverride.*.
The original system prompt Copilot Chat assembled and sent on the most recent request. Send a chat message first, then refresh to capture it. Copy it here as a starting point, then paste into the box below and switch the mode to "Replace" to fully rewrite it.
How your text below is combined with Copilot's system prompt.
Combined with the original prompt per the mode above. Use it to add stricter rules, change the assistant's identity, or fully replace the default prompt.
One JSON object per line to rewrite individual parts of the prompt — e.g. swap the safety block. Keys: find (required), replace, isRegex (true/false), flags (default "g"). Applied in order, after the combine step. Example: {"find":"Keep your answers short and impersonal.","replace":"Be thorough and detailed."}
Auto "Try Again" when a chat request fails before any content is streamed. 0 = off, -1 = infinite, or the max number of extra attempts. Output already streamed is never duplicated. Sets customcopilot.chatRetries.
Delay between automatic chat retries.
Optional random extra delay (0–this value) added before each retry. Spreads out retries so a batch of chats that failed together don't all hit the endpoint at once. 0 disables jitter. Sets customcopilot.chatRetryJitter.

Chat Generator

Write a prompt template with a placeholder, provide a list of values, and launch one Copilot chat per value with the mode and model you choose. Great for running the same task across many files, items or variants.
Use [REPLACE_THAT] (simple mode) or [KEY] tokens (advanced mode) as placeholders that get substituted for each generated chat.
Token replaced by each line in simple mode.
Simple = one value per line. Advanced = JSONL objects.
Each non-empty line generates one chat.
Copilot mode used for every generated chat.
Pick one of your configured models, or use the active one.
Sequential waits for each chat to finish (reliable). Parallel fires them with a delay so sessions run concurrently (best-effort — VS Code has no true parallel API).