feat(llm): native support for OpenAI, Anthropic, and Gemini#45
Open
saksenaaishwarya wants to merge 1 commit into
Open
feat(llm): native support for OpenAI, Anthropic, and Gemini#45saksenaaishwarya wants to merge 1 commit into
saksenaaishwarya wants to merge 1 commit into
Conversation
Introduces an LLMProvider contract and factory under src/llm so the CLI can target Ollama, OpenAI, Anthropic, or Gemini based on a new `backend` config field. API keys resolve from LLM_API_KEY first, then the provider-specific variable (OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY). When the selected backend is unavailable, behavior matches today's Ollama-down path and falls back to the rule-based engine. Existing public exports in src/llm/checkOllama and src/llm/ollamaEnhancer are preserved as thin re-exports so the gitbun-vscode extension and any external consumers keep compiling. Closes nirvik34#34
|
@saksenaaishwarya is attempting to deploy a commit to the nirvik34's projects Team on Vercel. A member of the Team first needs to authorize it. |
Contributor
|
Warning
|
| Layer / File(s) | Summary |
|---|---|
Type system and utilities src/llm/types.ts |
Introduces ProviderName, LLMProvider interface, ProviderConfig, and ProviderEnv types. Implements resolveApiKey() to handle generic LLM_API_KEY precedence over provider-specific keys. Exports COMMIT_SYSTEM_PROMPT, buildUserPrompt(), and cleanCommitOutput() helpers used by all providers. |
Provider factory pattern src/llm/index.ts, src/llm/factory.test.ts |
createProvider() factory dispatches to backend-specific provider creators (openai, anthropic, gemini, ollama) with exhaustiveness checking. Tests validate backend selection, API key gating for cloud providers, and isAvailable()/enhanceCommit() behavior. |
Cloud provider implementations src/llm/providers/openai.ts, src/llm/providers/anthropic.ts, src/llm/providers/gemini.ts, src/llm/providers/*.test.ts |
Each provider implements isAvailable(), resolveModel(), and enhanceCommit() using a 20s timeout, system/user prompts, and graceful fallback to original message on any error. Tests cover API key precedence, request formation, response parsing, and error paths including missing content and HTTP failures. |
Ollama provider refactor src/llm/providers/ollama.ts, src/llm/checkOllama.ts, src/llm/ollamaEnhancer.ts |
Ollama logic extracted into a provider module with functions for availability, model discovery (preferred-name matching with fallback), and commit enhancement. checkOllama.ts now re-exports from the provider module; ollamaEnhancer.ts delegates to enhanceCommitOllama(). |
Main entrypoint integration src/index.ts, README.md |
AI enhancement block replaced with provider-based flow: creates provider via factory, checks availability, resolves model from config/options, calls provider.enhanceCommit(), and falls back to rule-based engine on unavailability or exceptions. Documentation added describing all supported backends, environment variable setup, and fallback behavior. |
Estimated code review effort
🎯 4 (Complex) | ⏱️ ~60 minutes
Suggested labels
level: advanced, quality:clean, type:testing
Suggested reviewers
- 2PieRadian
Poem
🐰 Four backends now dance in harmony,
From OpenAI to Gemini with glee,
A factory dispatches with type-safe care,
Fallbacks ensure robustness everywhere,
Ollama still leads, but clouds have their share! ☁️✨
🚥 Pre-merge checks | ✅ 4 | ❌ 1
❌ Failed checks (1 warning)
| Check name | Status | Explanation | Resolution |
|---|---|---|---|
| Docstring Coverage | Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. | Write docstrings for the functions missing them to satisfy the coverage threshold. |
✅ Passed checks (4 passed)
| Check name | Status | Explanation |
|---|---|---|
| Description Check | ✅ Passed | Check skipped - CodeRabbit’s high-level summary is enabled. |
| Title check | ✅ Passed | The title clearly and concisely summarizes the main change: adding native support for three cloud LLM providers. |
| Linked Issues check | ✅ Passed | All coding objectives from issue #34 are met: modular provider interface, backend selection via config, API key resolution, soft-failure behavior, and backward compatibility with Ollama. |
| Out of Scope Changes check | ✅ Passed | All changes are directly aligned with issue #34's objectives; no unrelated modifications outside the provider abstraction, config, or documentation scope. |
✏️ Tip: You can configure your own custom pre-merge checks in the settings.
✨ Finishing Touches
🧪 Generate unit tests (beta)
- Create PR with unit tests
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 @coderabbitai help to get the list of available commands and usage tips.
![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: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/llm/providers/gemini.ts`:
- Around line 52-55: The request currently embeds apiKey into the request URL
(constructed via GEMINI_BASE_URL, model, and apiKey) and later logs the raw
error object on failures, risking credential leakage; change the request to
remove apiKey from the URL (keep GEMINI_BASE_URL and model encoding) and send
the key in a header (e.g., Authorization: Bearer <apiKey> or an appropriate
X-API-Key header) instead, and replace the raw error dump with a sanitized log
that includes only non-sensitive details (error.message, status code or a
generic failure message) and ensure any logged strings do not contain apiKey or
its encoded form.
In `@src/llm/providers/ollama.ts`:
- Around line 133-135: In resolveModel, the function currently returns the
user-provided requested value as-is which allows whitespace-only strings to
bypass model discovery; change the logic in resolveModel to trim requested
(e.g., const r = requested?.trim()) and only return it when r is a non-empty
string, otherwise call getBestModel(env) and return that (or null) so
whitespace-only input doesn't skip auto-selection.
In `@src/llm/types.ts`:
- Around line 61-65: cleanCommitOutput currently only strips a single wrapper
char and splits on "\n", which leaves triple-backtick fences and CRLF sequences
intact and also the function does not return the final value; update
cleanCommitOutput to first normalize line endings (convert CRLF to LF), then
trim whitespace, strip any surrounding fence or quote sequences (handle
one-or-more leading/trailing backticks or quotes and optional fenced-language
markers like ```js), remove a leading case-insensitive "commit:" token, then
take the first line (split on LF) and return that string; reference the
cleanCommitOutput function to locate and implement these changes.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: ca047072-a538-4712-91f8-33338404b743
📒 Files selected for processing (14)
README.mdsrc/index.tssrc/llm/checkOllama.tssrc/llm/factory.test.tssrc/llm/index.tssrc/llm/ollamaEnhancer.tssrc/llm/providers/anthropic.test.tssrc/llm/providers/anthropic.tssrc/llm/providers/gemini.test.tssrc/llm/providers/gemini.tssrc/llm/providers/ollama.tssrc/llm/providers/openai.test.tssrc/llm/providers/openai.tssrc/llm/types.ts
Comment on lines
+52
to
+55
| const url = `${GEMINI_BASE_URL}/${encodeURIComponent( | ||
| model | ||
| )}:generateContent?key=${encodeURIComponent(apiKey)}`; | ||
|
|
Contributor
There was a problem hiding this comment.
Avoid potential API-key leakage in error logs.
Line 52-Line 55 places the key in the URL, and Line 87 logs the raw error object. On request failures, this can leak credentials into logs. Log a sanitized message instead of dumping the full error object.
🔒 Proposed fix
- } catch (error) {
- console.log("\nAI Enhancement Failed:", error);
+ } catch (error) {
+ const reason = error instanceof Error ? error.name : "unknown";
+ console.log(`\nGemini request failed (${reason}).`);
return originalMessage;
} finally {Also applies to: 86-87
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/llm/providers/gemini.ts` around lines 52 - 55, The request currently
embeds apiKey into the request URL (constructed via GEMINI_BASE_URL, model, and
apiKey) and later logs the raw error object on failures, risking credential
leakage; change the request to remove apiKey from the URL (keep GEMINI_BASE_URL
and model encoding) and send the key in a header (e.g., Authorization: Bearer
<apiKey> or an appropriate X-API-Key header) instead, and replace the raw error
dump with a sanitized log that includes only non-sensitive details
(error.message, status code or a generic failure message) and ensure any logged
strings do not contain apiKey or its encoded form.
Comment on lines
+133
to
+135
| async resolveModel(requested?: string): Promise<string | null> { | ||
| if (requested) return requested; | ||
| const best = await getBestModel(env); |
Contributor
There was a problem hiding this comment.
Trim user-provided model before bypassing auto-selection.
Line 134 returns requested as-is. A whitespace-only value will skip discovery and send an invalid model downstream.
✅ Proposed fix
async resolveModel(requested?: string): Promise<string | null> {
- if (requested) return requested;
+ const trimmed = requested?.trim();
+ if (trimmed) return trimmed;
const best = await getBestModel(env);
return best ?? FALLBACK_MODEL;
},📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| async resolveModel(requested?: string): Promise<string | null> { | |
| if (requested) return requested; | |
| const best = await getBestModel(env); | |
| async resolveModel(requested?: string): Promise<string | null> { | |
| const trimmed = requested?.trim(); | |
| if (trimmed) return trimmed; | |
| const best = await getBestModel(env); | |
| return best ?? FALLBACK_MODEL; | |
| }, |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/llm/providers/ollama.ts` around lines 133 - 135, In resolveModel, the
function currently returns the user-provided requested value as-is which allows
whitespace-only strings to bypass model discovery; change the logic in
resolveModel to trim requested (e.g., const r = requested?.trim()) and only
return it when r is a non-empty string, otherwise call getBestModel(env) and
return that (or null) so whitespace-only input doesn't skip auto-selection.
Comment on lines
+61
to
+65
| export function cleanCommitOutput(raw: string): string { | ||
| let result = raw.trim(); | ||
| result = result.replace(/^[`"']|[`"']$/g, ""); | ||
| result = result.replace(/^commit:\s*/i, ""); | ||
| result = result.split("\n")[0]; |
Contributor
There was a problem hiding this comment.
Harden output cleaning for fenced/CRLF responses.
Current stripping removes only a single wrapper character and can leave malformed output for responses like triple backticks or CRLF lines.
Suggested patch
export function cleanCommitOutput(raw: string): string {
let result = raw.trim();
- result = result.replace(/^[`"']|[`"']$/g, "");
+ result = result.replace(/^[`"']+|[`"']+$/g, "");
result = result.replace(/^commit:\s*/i, "");
- result = result.split("\n")[0];
- return result;
+ result = result.split(/\r?\n/)[0]?.trim() ?? "";
+ return result;
}🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/llm/types.ts` around lines 61 - 65, cleanCommitOutput currently only
strips a single wrapper char and splits on "\n", which leaves triple-backtick
fences and CRLF sequences intact and also the function does not return the final
value; update cleanCommitOutput to first normalize line endings (convert CRLF to
LF), then trim whitespace, strip any surrounding fence or quote sequences
(handle one-or-more leading/trailing backticks or quotes and optional
fenced-language markers like ```js), remove a leading case-insensitive "commit:"
token, then take the first line (split on LF) and return that string; reference
the cleanCommitOutput function to locate and implement these changes.
7 tasks
nirvik34
requested changes
May 24, 2026
Owner
nirvik34
left a comment
nirvik34
left a comment
There was a problem hiding this comment.
Resolve the merge conflict
seciruty issue of gemini api... Move the API key to a header (e.g., Authorization: Bearer or X-API-Key) and remove it from the URL
In ollama.ts trim the requested value and only return it if non-empty; otherwise call getBestModel().
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.
Summary
Closes #34. Adds a modular
LLMProviderinterface undersrc/llm/so Gitbun can talk to OpenAI, Anthropic, and Gemini in addition to Ollama. Backend is picked via the existing config layer (.smartcommitrc/package.json#smartcommit); API keys come from the environment.Root cause / motivation
src/llm/was hardcoded to a local Ollama HTTP API. Contributors and CI users without a GPU asked for hosted-model support but the call sites insrc/index.tsimported Ollama-specific functions directly, so adding a backend would have required if/else trees in every call path.Solution
LLMProvidercontract insrc/llm/types.tswithisAvailable(),resolveModel(),enhanceCommit().src/llm/providers/(ollama.ts,openai.ts,anthropic.ts,gemini.ts), each using the existingnode-fetchdependency against the vendor's REST API — no SDK adds.createProvider(config, env)factory insrc/llm/index.tsselects the backend; default staysollamafor backwards compatibility.src/index.tsswapped to the factory; spinner / fallback messages are now provider-aware ("openai is not available. Using rule-based commit.").src/llm/checkOllama.tsandsrc/llm/ollamaEnhancer.tsreduced to thin re-exports so thegitbun-vscodeextension and any external imports keep compiling.Key changes
backendconfig field:ollama(default) |openai|anthropic|gemini.LLM_API_KEY→{PROVIDER}_API_KEY.gpt-4o-mini,claude-haiku-4-5-20251001,gemini-1.5-flash. Overridable with the existing--model <name>flag ormodelconfig key.Testing
npm run lintclean;tsc --noEmitclean.Backwards compatibility / risks
ollama; existing.smartcommitrcfiles behave identically.OLLAMA_HOSTenv var still honored.enhanceCommit,isOllamaRunning,getBestModel,getAvailableModels) kept at their original paths — the VSCode extension is unaffected.Checklist
Summary by CodeRabbit
New Features
LLM_API_KEYpreferredDocumentation