The Anatomy of a Feedback → PR Pipeline
90 Seconds from Feedback to Pull Request
A user clicks "Submit Feedback" in your app. 90 seconds later, a pull request lands on your repo — with the fix, a description linking to the original feedback, and green CI checks.
This is the core loop of SOLO. Let's walk through every step.
Step 1: Widget Capture (0s)
The journey starts with the feedback widget — a lightweight JavaScript snippet embedded in your app.
When a user submits feedback, the widget captures:
The payload is sent to SOLO's API as a POST request with the tenant token (a unique ID for your repo).
Step 2: Ingestion and Storage (0.5s)
The submission lands in the feedback_submissions table, linked to your tenant. It's immediately visible in your dashboard under the Inbox tab.
At this point, the submission is in "pending" status — waiting for triage.
Step 3: AI Triage (1-2s)
The triage system kicks in asynchronously. It reads the submission and produces:
The submission status moves from "pending" to "triaged."
Step 4: Agent Dispatch (2s)
If the submission scores above the auto-fix threshold (configurable, default 80), the coding agent is dispatched automatically. On lower scores, you can trigger the agent manually from the dashboard.
The dispatch process:
The agent uses the LLM configured for your tenant — either the platform Gemini key (Free plan) or your own BYOK key (Starter/Pro).
Step 5: Code Generation (30-60s)
This is where the magic happens. The agent:
The agent is constrained to only modify files related to the feedback. No scope creep, no drive-by refactors.
Step 6: Build Verification (15-30s)
Before opening the PR, the agent runs your CI checks:
If any check fails, the agent reads the error output, fixes its code, and tries again. This loop runs up to 3 times (configurable).
Only PRs that pass all checks are opened. This is non-negotiable — we never ship broken code.
Step 7: PR Creation (2s)
The agent opens a pull request on your repo with:
The submission status updates to "implemented" and the PR appears in your dashboard under Pull Requests.
Step 8: Human Review
This is where you come in. The PR is ready for review — read the description, check the diff, and merge if it looks good.
Most PRs require no changes. Some need minor tweaks. A few get closed if the agent misunderstood the feedback.
The Full Timeline
Under 90 seconds, end to end. Compare that to the traditional flow: user reports bug, PM writes ticket, ticket sits in backlog, engineer picks it up, engineer context-switches, engineer writes fix, engineer runs tests, engineer opens PR. That's 3-5 business days on average.
Monitoring the Pipeline
Every step is visible in your SOLO dashboard:
You can also connect Slack or Discord to get notifications when PRs are opened or when high-priority feedback arrives.
Getting Started
The pipeline activates automatically. No configuration needed for basic usage.
SOLO turns user feedback into merged PRs. 90 seconds. Automatically.
- Feedback text — What the user typed
- Page URL — Where they were in the app
- Metadata — Browser, viewport, OS, timestamp
- Session context — Optional custom data you attach (user ID, plan, feature flags)
- Category: Bug, feature request, UX issue, performance, or praise
- Priority score: 0-100, weighted by severity, affected area, and user context
- Triage reasoning: A one-paragraph explanation of why it scored this way
- Affected files: Best-guess list of files that need changes (based on page URL and keywords)
- Duplicate check: Semantic similarity against recent submissions
- Linting (ESLint, Prettier, etc.)
- Type checking (TypeScript)
- Unit tests
- Integration tests (if configured)
- Title: A concise description of the fix
- Body: Links to the original feedback submission, a summary of changes, and the triage analysis
- Branch: A clean branch name like solo/fix-signup-button-safari
- Labels: Auto-applied based on the triage category
- Files changed: The minimal set of changes needed
- Inbox — Raw submissions with triage scores
- Pull Requests — Agent-generated PRs with status tracking
- Agent Jobs — Real-time view of running agent tasks
- Clusters — Auto-grouped similar feedback submissions