Skip to main content
How Tembo Works Refine, iterate, and improve on Tembo’s pull requests using natural language feedback directly in your Git provider.

​
Prerequisites

The Feedback Loop works automatically once you have:
  • GitHub integration or GitLab integration enabled in your Tembo workspace
  • At least one pull request or merge request created by Tembo (from error detection, manual assignment, etc.)
That’s it! No additional setup required.
Git Provider Support: The Feedback Loop currently supports GitHub and GitLab. Support for Bitbucket is coming soon!

​
How it works

  1. Initial Solution - Tembo creates a pull request (GitHub) or merge request (GitLab) to solve an issue
  2. Review & Comment - You review the PR/MR and leave feedback (tag @tembo in your review or comments)
  3. Automatic Retry - Tembo processes your feedback and updates the PR/MR with improvements
  4. Iteration - Continue providing feedback until you’re satisfied with the solution
Credit Usage: Processing feedback through the Feedback Loop consumes credits based on the complexity of the requested changes and the size of the codebase context. Each feedback iteration is treated as a separate task. Learn more about credit consumption.
Important: You must mention @tembo in your PR/MR comments or reviews for Tembo to process your feedback. This prevents accidental triggers from internal team discussions or comments meant for other team members.

​
Providing Feedback

​
Option 1: Use PR/MR Review function

GitHub Review Interface
  • Submit a formal review using your Git provider’s built-in review feature (GitHub PR reviews or GitLab MR reviews)
  • Tag “@tembo” in your review summary or in any specific comments you want Tembo to act on (untagged feedback is ignored)
  • Advantages:
    • Tembo knows exactly which lines you’re commenting on for precise fixes
    • You can batch multiple feedback points in a single review
    • Cleaner conversation thread with organized feedback
    • Standard workflow that integrates with your team’s process

​
Option 2: Add comments

  • Leave regular comments in the PR/MR conversation
  • You must mention “@tembo” in the comment to trigger processing (untagged comments are ignored)
Comment Review Example

​
Example Workflow

Scenario: Tembo creates a PR with a basic try-catch for database timeouts. You review and comment “@tembo add retry logic with exponential backoff, logging, and unit tests.” Tembo processes your feedback, pushes a new commit with all changes, and you continue iterating until satisfied.

​
What to Expect

Response Time:
  • Tembo typically acknowledges feedback within 1-2 minutes (👀 reaction)
  • Processing and implementing changes usually takes 3-10 minutes depending on complexity
  • You’ll see new commits appear in the PR once changes are complete
Processing Indicators:
  • 👀 = Tembo has seen your feedback and is analyzing it
  • 🚀 = Tembo is actively working on implementing changes
  • New commits = Changes are complete and ready for your review

​
Best Practices

​
Writing Effective Feedback

✅ Be specific and actionable: 
@tembo Add input validation for the email field using regex
@tembo Implement retry logic with exponential backoff for API calls
❌ Too vague: 
@tembo fix the validation
@tembo make it better
✅ Batch feedback in a single review:
  • Add multiple comments to different lines in one review (GitHub PR review or GitLab MR review)
  • Tag “@tembo” on the review summary or individual comments you want addressed; Tembo processes all tagged points together for cohesive improvements
  • Cleaner conversation thread and better context awareness
ℹ️ Tagging is required:
  • Tembo only acts on PR/MR feedback that mentions “@tembo” (in reviews or regular comments)
  • This prevents accidental triggers from internal team discussions or other bot messages

​
What Tembo’s Feedback Loop Handles Well

  • Code improvements: Performance, error handling, refactoring, type safety
  • Testing and documentation: Unit tests, code comments, README updates
  • Security and bugs: Vulnerability fixes, edge cases, error handling
  • Alignment: Changes that follow your existing code patterns

​
Limitations

  • Complex architectural changes may need multiple iterations
  • Very large-scope changes are better as separate issues
  • Works best on code within the current PR scope
You are responsible for all merged code. Read our AI First Software Development guide for more information.