feat(llm): add GitHub Copilot as a first-class LLM provider#394
Merged
jamiepine merged 2 commits intospacedriveapp:mainfrom Mar 12, 2026
Merged
feat(llm): add GitHub Copilot as a first-class LLM provider#394jamiepine merged 2 commits intospacedriveapp:mainfrom
jamiepine merged 2 commits intospacedriveapp:mainfrom
Conversation
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
WalkthroughAdds GitHub Copilot as a new LLM provider across frontend and backend: API/schema additions, UI labels/icons/settings, config/TOML/env support, provider routing, a new auth module for PAT→Copilot token exchange and caching, and LlmManager integration for token lifecycle and provider construction. Changes
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
Suggested reviewers
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
ibhagwan
force-pushed
the
spacebot-copilot
branch
2 times, most recently
from
a198f35 to
d84fe73
Compare
Contributor
Author
|
@coderabbitai review |
Contributor
✅ Actions performedReview triggered.
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 4
🧹 Nitpick comments (5)
src/github_copilot_auth.rs (2)
63-75: Creates new HTTP client per token exchange.
exchange_github_tokencreates a newreqwest::Clienton each call. While this works, it doesn't benefit from connection pooling. Consider accepting theLlmManager's HTTP client as a parameter, or document why a fresh client is preferred here.♻️ Proposed signature change to reuse HTTP client
-pub async fn exchange_github_token(github_pat: &str, pat_hash: String) -> Result<CopilotToken> { - let client = reqwest::Client::new(); +pub async fn exchange_github_token( + client: &reqwest::Client, + github_pat: &str, + pat_hash: String, +) -> Result<CopilotToken> {Then update the caller in
src/llm/manager.rsto passself.http_client().🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/github_copilot_auth.rs` around lines 63 - 75, The exchange_github_token function currently constructs a new reqwest::Client on each call (in exchange_github_token), causing missed connection pooling; change the signature of exchange_github_token to accept an existing HTTP client (e.g., &reqwest::Client or &Client) instead of creating one inside the function, remove the Client::new() call, and use the passed client for the request; then update callers (notably in LlmManager or src/llm/manager.rs) to forward self.http_client() when invoking exchange_github_token so the shared client is reused.
176-191: File write before permission set creates brief window of exposure.
save_cached_tokenwrites the file first, then sets restrictive permissions. On Unix systems, there's a brief window where the file may be readable by other users (depending on umask). Consider usingOpenOptionswith mode set at creation time.🔒 Proposed fix for atomic secure file creation on Unix
pub fn save_cached_token(instance_dir: &Path, token: &CopilotToken) -> Result<()> { let path = credentials_path(instance_dir); let data = serde_json::to_string_pretty(token).context("failed to serialize Copilot token")?; - std::fs::write(&path, &data).with_context(|| format!("failed to write {}", path.display()))?; - #[cfg(unix)] { - use std::os::unix::fs::PermissionsExt; - std::fs::set_permissions(&path, std::fs::Permissions::from_mode(0o600)) - .with_context(|| format!("failed to set permissions on {}", path.display()))?; + use std::fs::OpenOptions; + use std::io::Write; + use std::os::unix::fs::OpenOptionsExt; + let mut file = OpenOptions::new() + .write(true) + .create(true) + .truncate(true) + .mode(0o600) + .open(&path) + .with_context(|| format!("failed to create {}", path.display()))?; + file.write_all(data.as_bytes()) + .with_context(|| format!("failed to write {}", path.display()))?; } + + #[cfg(not(unix))] + { + std::fs::write(&path, &data).with_context(|| format!("failed to write {}", path.display()))?; + } Ok(()) }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/github_copilot_auth.rs` around lines 176 - 191, The current save_cached_token writes the file then sets permissions, leaving a brief exposure window; modify save_cached_token to create-and-write the file with restrictive mode atomically on Unix by using std::fs::OpenOptions with std::os::unix::fs::OpenOptionsExt::mode(0o600) (use create(true)|truncate(true)|write(true)), write the serde_json data into that open file (and optionally flush/fsync), and remove the separate std::fs::set_permissions call; keep the existing serialize to data and the credentials_path lookup, and retain the non-Unix path behavior (or fall back to current write+set_permissions only on non-Unix platforms).src/llm/manager.rs (2)
70-73: Inconsistent error handling betweenset_instance_dirandwith_instance_dir.
set_instance_dirsilently discards load errors viaif let Ok(Some(...)), whilewith_instance_dir(lines 109-119) logs warnings. For consistency and debuggability, consider logging failures here as well.♻️ Proposed fix for consistent error handling
- if let Ok(Some(token)) = crate::github_copilot_auth::load_cached_token(&instance_dir) { - tracing::info!("loaded GitHub Copilot token from github_copilot_token.json"); - *self.copilot_token.write().await = Some(token); - } + match crate::github_copilot_auth::load_cached_token(&instance_dir) { + Ok(Some(token)) => { + tracing::info!("loaded GitHub Copilot token from github_copilot_token.json"); + *self.copilot_token.write().await = Some(token); + } + Ok(None) => {} + Err(error) => { + tracing::warn!(%error, "failed to load GitHub Copilot token"); + } + }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/llm/manager.rs` around lines 70 - 73, The current set_instance_dir code silently ignores errors when calling crate::github_copilot_auth::load_cached_token(&instance_dir); update it to mirror with_instance_dir's behavior by matching the Result and logging warnings on Err(e) and when Ok(None) to aid debugging, while still setting *self.copilot_token.write().await = Some(token) on Ok(Some(token)); reference the load_cached_token call and self.copilot_token mutation to locate and modify the logic.
369-378: Hardcoded editor version strings may become stale.The
editor-versionandeditor-plugin-versionheaders use hardcoded values (vscode/1.96.2,copilot-chat/0.26.7). These may need periodic updates as GitHub Copilot evolves its API requirements. Consider extracting these as constants or making them configurable.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/llm/manager.rs` around lines 369 - 378, The extra_headers block currently hardcodes "editor-version" and "editor-plugin-version" values (inside the extra_headers vector) which will become stale; refactor by replacing those hardcoded strings with constants or configuration-driven values (e.g., EDITOR_VERSION and EDITOR_PLUGIN_VERSION constants or env/config fields) and update the code that builds extra_headers to read those constants/config values so they can be updated without editing the header literal in extra_headers.src/config/load.rs (1)
434-438: Environment variable fallback chain may unintentionally use general-purpose GitHub tokens.The fallback chain includes
GH_TOKENandGITHUB_TOKEN, which are commonly set for general GitHub CLI/API operations. These tokens may not have Copilot access, leading to confusing errors at runtime. Consider:
- Documenting this behavior clearly
- Only falling back to these if
GITHUB_COPILOT_API_KEYandCOPILOT_GITHUB_TOKENare not setThe current implementation already does this correctly (chained
or_else), but users should be warned that theirGH_TOKENwill be tried for Copilot access.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/config/load.rs` around lines 434 - 438, The fallback chain for github_copilot_key (the github_copilot_key field in load.rs) may unexpectedly pick up general-purpose GH_TOKEN or GITHUB_TOKEN; keep the existing or_else order but add a short code comment above github_copilot_key explaining that GH_TOKEN/GITHUB_TOKEN are only used as last-resort fallbacks and may not grant Copilot access, and also emit a runtime warning/log when the resolved github_copilot_key value originated from GH_TOKEN or GITHUB_TOKEN (e.g., detect the env var used during construction inside the same load/config initialization and call the project logger to warn the user).
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@interface/src/routes/Settings.tsx`:
- Around line 245-252: The Copilot provider card seeds the wrong default model
via provider.defaultModel which initializes modelInput and then gets applied to
every routing role on save; update the provider entry with id "github-copilot"
to use the correct defaultModel "github-copilot/gpt-4.1-mini" (or remove
defaultModel) and/or change the Settings save logic that uses modelInput to
avoid overwriting provider-specific routing defaults—only apply modelInput to
roles that have no existing model override (check the save handler that reads
modelInput and assigns to routing roles) so Copilot-specific routing defaults
are preserved.
In `@src/api/providers.rs`:
- Around line 420-421: get_providers currently uses
has_value("github_copilot_key", "GITHUB_COPILOT_API_KEY") which is inconsistent
with load_from_env's fallback list; update get_providers to consider the same
env keys used by load_from_env by checking for any of COPILOT_GITHUB_TOKEN,
GH_TOKEN, GITHUB_TOKEN in addition to GITHUB_COPILOT_API_KEY (and mirror this
change where get_providers checks copilot again, e.g., the second occurrence
around the provider status logic) so the UI reflects the same configured state
as load_from_env.
In `@src/config/providers.rs`:
- Around line 217-224: The provider test flow is incorrectly using
default_provider_config() for "github-copilot" and sending the raw PAT; update
the test path (where build_test_llm_config() / test_provider_model handle
providers) to special-case the "github-copilot" provider: call the GitHub
token-exchange flow (github_copilot_auth::exchange_github_token()) or reuse
LlmManager::get_github_copilot_provider() logic to exchange the PAT for a
Copilot token, set use_bearer_auth to true, set api_key to the exchanged token,
and derive base_url from the exchanged token's proxy-ep instead of using
GITHUB_COPILOT_DEFAULT_BASE_URL; do not call default_provider_config() for
"github-copilot" in the test path.
In `@src/config/types.rs`:
- Around line 509-513: The SecretField list currently only registers the
"github_copilot_key" -> "GITHUB_COPILOT_API_KEY" mapping; add additional
SecretField entries so SystemSecrets will resolve the other documented Copilot
env aliases (e.g. secret_name values "COPILOT_GITHUB_TOKEN", "GH_TOKEN", and
"GITHUB_TOKEN" with the same toml_key "github_copilot_key" or appropriate
toml_key variants) alongside the existing SecretField, so lookups for Copilot
honor all those environment variable names; update the entries near the existing
SecretField block to include these aliases.
---
Nitpick comments:
In `@src/config/load.rs`:
- Around line 434-438: The fallback chain for github_copilot_key (the
github_copilot_key field in load.rs) may unexpectedly pick up general-purpose
GH_TOKEN or GITHUB_TOKEN; keep the existing or_else order but add a short code
comment above github_copilot_key explaining that GH_TOKEN/GITHUB_TOKEN are only
used as last-resort fallbacks and may not grant Copilot access, and also emit a
runtime warning/log when the resolved github_copilot_key value originated from
GH_TOKEN or GITHUB_TOKEN (e.g., detect the env var used during construction
inside the same load/config initialization and call the project logger to warn
the user).
In `@src/github_copilot_auth.rs`:
- Around line 63-75: The exchange_github_token function currently constructs a
new reqwest::Client on each call (in exchange_github_token), causing missed
connection pooling; change the signature of exchange_github_token to accept an
existing HTTP client (e.g., &reqwest::Client or &Client) instead of creating one
inside the function, remove the Client::new() call, and use the passed client
for the request; then update callers (notably in LlmManager or
src/llm/manager.rs) to forward self.http_client() when invoking
exchange_github_token so the shared client is reused.
- Around line 176-191: The current save_cached_token writes the file then sets
permissions, leaving a brief exposure window; modify save_cached_token to
create-and-write the file with restrictive mode atomically on Unix by using
std::fs::OpenOptions with std::os::unix::fs::OpenOptionsExt::mode(0o600) (use
create(true)|truncate(true)|write(true)), write the serde_json data into that
open file (and optionally flush/fsync), and remove the separate
std::fs::set_permissions call; keep the existing serialize to data and the
credentials_path lookup, and retain the non-Unix path behavior (or fall back to
current write+set_permissions only on non-Unix platforms).
In `@src/llm/manager.rs`:
- Around line 70-73: The current set_instance_dir code silently ignores errors
when calling crate::github_copilot_auth::load_cached_token(&instance_dir);
update it to mirror with_instance_dir's behavior by matching the Result and
logging warnings on Err(e) and when Ok(None) to aid debugging, while still
setting *self.copilot_token.write().await = Some(token) on Ok(Some(token));
reference the load_cached_token call and self.copilot_token mutation to locate
and modify the logic.
- Around line 369-378: The extra_headers block currently hardcodes
"editor-version" and "editor-plugin-version" values (inside the extra_headers
vector) which will become stale; refactor by replacing those hardcoded strings
with constants or configuration-driven values (e.g., EDITOR_VERSION and
EDITOR_PLUGIN_VERSION constants or env/config fields) and update the code that
builds extra_headers to read those constants/config values so they can be
updated without editing the header literal in extra_headers.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 48d60412-3598-40c2-b4e4-0d16d1883433
📒 Files selected for processing (15)
interface/src/api/client.tsinterface/src/components/ModelSelect.tsxinterface/src/lib/providerIcons.tsxinterface/src/routes/Settings.tsxsrc/api/providers.rssrc/config/load.rssrc/config/providers.rssrc/config/toml_schema.rssrc/config/types.rssrc/github_copilot_auth.rssrc/lib.rssrc/llm/manager.rssrc/llm/model.rssrc/llm/routing.rssrc/tools/send_message_to_another_channel.rs
ibhagwan
force-pushed
the
spacebot-copilot
branch
from
ebd871f to
b7e7be7
Compare
Contributor
Author
|
@coderabbitai review |
Contributor
✅ Actions performedReview triggered.
|
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@src/llm/manager.rs`:
- Line 386: Update the shorthand GitHub Copilot provider registration to use the
chat completions API type: change the provider's ApiType from
ApiType::OpenAiCompletions to ApiType::OpenAiChatCompletions so the runtime and
shorthand configurations match and the provider uses the /chat/completions
endpoint; modify the registration code that constructs the provider (the
shorthand provider registration in load configuration) to set api_type =
ApiType::OpenAiChatCompletions, ensuring any related enum usages or matches
expect the chat variant.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 77fa6b34-af68-4941-8aee-03d977cb19ce
📒 Files selected for processing (4)
src/config/load.rssrc/config/providers.rssrc/github_copilot_auth.rssrc/llm/manager.rs
🚧 Files skipped from review as they are similar to previous changes (1)
- src/config/load.rs
ibhagwan
force-pushed
the
spacebot-copilot
branch
from
b7e7be7 to
771504e
Compare
Contributor
Author
|
@coderabbitai review |
Contributor
✅ Actions performedReview triggered.
|
![tembo[bot]](https://avatars.githubusercontent.com/in/1223475?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
🧹 Nitpick comments (1)
src/github_copilot_auth.rs (1)
114-129: Consider documenting the heuristic threshold.The 10 billion threshold for distinguishing seconds from milliseconds works for dates through ~2286 (in seconds) or ~1970+116 days (in milliseconds). This is reasonable, but a brief inline comment explaining the boundary would help future maintainers.
📝 Suggested documentation
fn parse_expires_at(value: &serde_json::Value) -> Option<i64> { let raw = match value { serde_json::Value::Number(number) => number.as_i64()?, serde_json::Value::String(string) => string.trim().parse::<i64>().ok()?, _ => return None, }; + // Heuristic: 10^10 seconds ≈ year 2286, so values above this are likely milliseconds. + // This safely handles both formats for any reasonable token expiry. if raw > 10_000_000_000 { Some(raw) // already milliseconds } else { Some(raw * 1000) // convert seconds to milliseconds } }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/github_copilot_auth.rs` around lines 114 - 129, Add a concise inline comment in parse_expires_at explaining the 10_000_000_000 heuristic: state that values >10 billion are treated as milliseconds because 10_000_000_000 seconds corresponds to a date around year 2286, so smaller values are likely seconds and are converted to milliseconds by multiplying by 1000; reference the function name parse_expires_at and place the comment near the threshold check to make the rationale clear for future maintainers.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Nitpick comments:
In `@src/github_copilot_auth.rs`:
- Around line 114-129: Add a concise inline comment in parse_expires_at
explaining the 10_000_000_000 heuristic: state that values >10 billion are
treated as milliseconds because 10_000_000_000 seconds corresponds to a date
around year 2286, so smaller values are likely seconds and are converted to
milliseconds by multiplying by 1000; reference the function name
parse_expires_at and place the comment near the threshold check to make the
rationale clear for future maintainers.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 17c39d16-9859-4f41-b116-e5a2e89f95bc
📒 Files selected for processing (2)
src/github_copilot_auth.rssrc/llm/manager.rs
ibhagwan
force-pushed
the
spacebot-copilot
branch
from
d62a5a0 to
29a7ae3
Compare
Contributor
Author
|
@coderabbitai review |
Contributor
✅ Actions performedReview triggered.
|
Contributor
There was a problem hiding this comment.
🧹 Nitpick comments (1)
src/github_copilot_auth.rs (1)
234-305: Good test coverage for core functionality.The tests cover the main paths including expiry parsing, hashing determinism, and URL derivation.
Consider adding a test for the security validation (rejecting hostnames that don't end with
.githubcopilot.com):🧪 Optional test for security validation
#[test] fn derive_base_url_rejects_invalid_suffix() { let token = "tid=abc123;proxy-ep=proxy.evil.com;st=dotcom"; assert_eq!(derive_base_url_from_token(token), None); }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/github_copilot_auth.rs` around lines 234 - 305, Add a unit test to cover the security validation that rejects proxy endpoints not ending with ".githubcopilot.com": create a test named like derive_base_url_rejects_invalid_suffix that calls derive_base_url_from_token with a token containing "proxy-ep=proxy.evil.com" (and other fields like tid/st) and asserts the result is None; this ensures the hostname suffix check in derive_base_url_from_token is enforced.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Nitpick comments:
In `@src/github_copilot_auth.rs`:
- Around line 234-305: Add a unit test to cover the security validation that
rejects proxy endpoints not ending with ".githubcopilot.com": create a test
named like derive_base_url_rejects_invalid_suffix that calls
derive_base_url_from_token with a token containing "proxy-ep=proxy.evil.com"
(and other fields like tid/st) and asserts the result is None; this ensures the
hostname suffix check in derive_base_url_from_token is enforced.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: a28145af-97e7-49dc-b1eb-1efa4f8d84b0
📒 Files selected for processing (2)
src/github_copilot_auth.rssrc/llm/manager.rs
ibhagwan
force-pushed
the
spacebot-copilot
branch
from
29a7ae3 to
4e752b4
Compare
Contributor
There was a problem hiding this comment.
🧹 Nitpick comments (2)
src/llm/manager.rs (2)
22-28: Hardcoded editor/plugin versions will become stale over time.These version strings mimic a specific VSCode/Copilot Chat release. When GitHub updates their API or deprecates older client versions, requests may start failing. Consider making these configurable or documenting the update process.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/llm/manager.rs` around lines 22 - 28, The hardcoded constants COPILOT_EDITOR_VERSION and COPILOT_EDITOR_PLUGIN_VERSION can become stale; change them to configurable values (e.g., read from environment variables or a config struct) with sensible defaults equal to the current constants, update any places that reference these constants to use the new config fields or accessor functions, and add documentation/comments indicating where to update the defaults when GitHub/Copilot versions change so the values can be rotated without code edits.
412-415: Consider clarifying that the disk cache is not deleted.This method only clears the in-memory token. The on-disk cache file persists and will be reloaded on the next manager initialization. This is fine if
delete_provideralways handles file deletion first (as shown insrc/api/providers.rs:929-934), but the docstring could be more explicit to prevent misuse.📝 Suggested docstring clarification
- /// Clear cached GitHub Copilot token from memory. + /// Clear cached GitHub Copilot token from memory only. + /// + /// Note: Does not delete the on-disk cache file. Use + /// `github_copilot_auth::credentials_path()` and delete the file separately + /// if persistent removal is needed (e.g., in `delete_provider`). pub async fn clear_copilot_token(&self) {🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/llm/manager.rs` around lines 412 - 415, Update the docstring for the method clear_copilot_token to explicitly state it only clears the in-memory Copilot token and does not delete any on-disk cache file (the on-disk token will persist and be reloaded on next Manager initialization); mention that callers should use delete_provider (or an equivalent file-deleting routine) if they need the disk cache removed as well.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Nitpick comments:
In `@src/llm/manager.rs`:
- Around line 22-28: The hardcoded constants COPILOT_EDITOR_VERSION and
COPILOT_EDITOR_PLUGIN_VERSION can become stale; change them to configurable
values (e.g., read from environment variables or a config struct) with sensible
defaults equal to the current constants, update any places that reference these
constants to use the new config fields or accessor functions, and add
documentation/comments indicating where to update the defaults when
GitHub/Copilot versions change so the values can be rotated without code edits.
- Around line 412-415: Update the docstring for the method clear_copilot_token
to explicitly state it only clears the in-memory Copilot token and does not
delete any on-disk cache file (the on-disk token will persist and be reloaded on
next Manager initialization); mention that callers should use delete_provider
(or an equivalent file-deleting routine) if they need the disk cache removed as
well.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 57353777-d27f-4f5f-a683-b43041700dd4
📒 Files selected for processing (2)
src/github_copilot_auth.rssrc/llm/manager.rs
Two-step auth: exchanges a GitHub PAT for a short-lived Copilot API token, caches it to disk, and auto-refreshes when expired. The Copilot endpoint is OpenAI-compatible so it routes through ApiType::OpenAiCompletions. Backend: - New github_copilot_auth module: token exchange, disk caching (0600), base URL derivation from proxy-ep field, expiry handling - Config: github_copilot_key in [llm] section (TOML + env:GITHUB_COPILOT_API_KEY) - LlmManager: copilot_token RwLock, lazy exchange/refresh, provider config generation with derived base URL and bearer auth - Routing: defaults (channel=claude-sonnet-4, worker=gpt-4.1-mini), prefix mapping, priority ordering - API: provider status, TOML key mapping, delete handler (clears cached token file + in-memory state) Frontend: - Settings page: add/update/remove provider entry - ModelSelect: provider label and ordering - Provider icons: GithubCopilot from @lobehub/icons
ibhagwan
force-pushed
the
spacebot-copilot
branch
from
4e752b4 to
8a82586
Compare
jamiepine
approved these changes
Mar 12, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two-step auth: exchanges a GitHub PAT for a short-lived Copilot API
token, caches it to disk, and auto-refreshes when expired. The Copilot
endpoint is OpenAI-compatible so it routes through ApiType::OpenAiCompletions.
Backend:
base URL derivation from proxy-ep field, expiry handling
GITHUB_COPILOT_API_KEY, COPILOT_GITHUB_TOKEN, GH_TOKEN, GITHUB_TOKEN)
config generation with derived base URL and bearer auth
prefix mapping, priority ordering
token file + in-memory state)
Frontend:
Note
This PR implements GitHub Copilot as a first-class LLM provider with automatic token exchange and refresh. The backend handles GitHub PAT → Copilot API token conversion, caches tokens to disk (restricted 0600 permissions), and auto-refreshes expired tokens. Configuration via GITHUB_COPILOT_API_KEY environment variable with TOML fallback. Frontend adds provider settings UI, model labeling, and icon support. The Copilot endpoint (OpenAI-compatible) derives its base URL from the token's proxy-ep field with fallback to default endpoint. Latest improvements: simplified env var handling, atomic file creation on Unix, proper error logging, and hardcoded editor version headers for Copilot API compliance.
Written by Tembo for commit 771504e.