User Guide

Everything you need to ship with VibeAssist.

From your first app idea to a clean, AI-maintained backlog wired into the tools you already use. Pick a topic below — or search for what you need.

Browse by topic

Getting started

Get access, land in the dashboard, and ship your first idea in under five minutes.

Read

Projects & app ideas

Every project is an isolated workspace with its own backlog, knowledge base, and integrations.

Read

The backlog

Epics group themes. Tasks are the atomic units. Statuses move work from idea to shipped.

Read

AI task generation

Generate tasks from an idea, then let the AI keep the backlog tidy — without making any changes you didn't approve.

Read

AI provider settings

Bring your own key from any supported provider. We store it encrypted and use it only for your projects.

Read

GitHub integration

Link a repo, then let VibeAssist detect completed work straight from your commits.

Read

GitHub webhook (AI ghost sync)

Let GitHub tell VibeAssist when code lands or issues close, so your backlog updates itself.

Read

Knowledge base

Persistent context — architecture, personas, brand, decisions — that keeps the AI grounded.

Read

Sidebar map

A guided tour of every group and link in the left sidebar — what each one does and when to reach for it.

Read

Coding sessions

Plan a focused work block, log the prompts and lessons as you go, and export the result as markdown.

Read

Project configuration

Vibe Templates, architectural constraints, and the design patterns and libraries that drive the Cheat Sheet.

Read

Project cheat sheet

A read-only, at-a-glance reference of the design patterns, libraries and constraints chosen for this project.

Read

UI consistency review

Scan the linked repo for UI pattern deviations and approve AI-proposed fixes.

Read

AI usage

Per-project breakdown of token consumption and estimated cost across every AI model you've used.

Read

Tips for power users

Small habits that compound: keep context fresh, scan often, trust your reviews.

Read

Troubleshooting & FAQ

The most common gotchas and how to clear them in under a minute.

Read
Get started

Getting started

Get access, land in the dashboard, and ship your first idea in under five minutes.

VibeAssist is a workspace that turns rough product ideas into clean, AI-ready backlogs — then helps you ship them with whatever coding tool you love (Lovable, Bolt, Cursor, Replit, Claude Code, you name it).

  1. 1

    Get access

    Subscribe from pricing or request Free Beta access. Once approved or paid, use the emailed sign-in link to reach your dashboard.

  2. 2

    Create your first project

    Click New Project, paste your app idea in plain English, and save. One paragraph is plenty.

  3. 3

    Generate a backlog

    Open the AI panel and ask it to draft tasks from your idea. Accept the ones you like — reject the rest.

Write like you'd brief a teammate

Mention the user, the core flow, and the constraint. "A timer app for sales reps that tracks calls per account" beats "a timer app" every time.

Get started

Projects & app ideas

Every project is an isolated workspace with its own backlog, knowledge base, and integrations.

Projects are the unit of organization in VibeAssist. Each one keeps its tasks, knowledge documents, GitHub link, and AI settings separate. You can have as many as you want and switch between them from the sidebar.

What lives inside a project

  • App idea. The original brief — editable any time.
  • Backlog. Epics and tasks, the heart of the workspace.
  • Knowledge base. Persistent context the AI reads from.
  • Integrations. GitHub repo link, webhook endpoints, etc.
Core workflow

The backlog

Epics group themes. Tasks are the atomic units. Statuses move work from idea to shipped.

Each project's backlog is organized into Epics (broad themes) and Tasks (atomic units of work). Click any task to open the edit dialog — change the title, description, status, or add implementation notes like commit URLs or decisions.

Task statuses

BacklogNot started yet.
In progressActively being worked on.
Pending reviewWaiting for a check.
CompletedWork is done.
ApprovedReviewed and accepted.
RejectedSent back for changes.

Notes are your future self's best friend

When a task ships, paste the commit URL and any gotchas into the notes field. Three months later you'll thank yourself.

AI & automation

AI task generation

Generate tasks from an idea, then let the AI keep the backlog tidy — without making any changes you didn't approve.

VibeAssist's LLM sees what's already in your backlog before suggesting anything, so it avoids duplicates and respects the work you've done. Every AI suggestion arrives as a proposal — nothing changes until you accept it.

Tidy your backlog

  • Merge proposals. Combine overlapping tasks into one.
  • Delete proposals. Retire tasks that are no longer relevant.
  • Re-prioritization. Suggested ordering based on dependencies.

You're always in control

Reject anything that doesn't fit. Your backlog stays exactly the way you want it.

AI & automation

AI provider settings

Bring your own key from any supported provider. We store it encrypted and use it only for your projects.

VibeAssist is BYOK (bring-your-own-key). Open Settings → AI Provider, pick a provider, paste your key and choose a default model. You can change provider or model any time — your data stays put.

Supported providers & models

Every provider below is wired into the same AI features (backlog generation, task cleanup, scan repository, prompt drafting, knowledge import, etc.). Pick whatever you already pay for.

  • OpenAI. gpt-4o-mini · gpt-4o · gpt-4.1-mini · gpt-4.1 · o4-mini. Best all-rounder for prompt drafting and task generation. Requires a paid OpenAI account (no free tier for API).
  • Anthropic (Claude). claude-3-5-sonnet · claude-3-5-haiku · claude-3-7-sonnet · claude-opus-4 · claude-sonnet-4. Strongest at long-form rewriting and code review style outputs. Pay-as-you-go billing only.
  • Google (Gemini). gemini-2.5-pro · gemini-2.5-flash · gemini-2.5-flash-lite. Free tier works for testing; check Google's data-use terms before sending sensitive content on the free tier.
  1. 1

    Get a key

    Each provider links straight to its API-keys page from the picker in Settings.

  2. 2

    Paste & save

    Drop it into the AI Provider section in Settings. It's encrypted at rest and never leaves your project.

  3. 3

    Pick a default model

    The dropdown filters to models your chosen provider supports.

Limits to expect

Every AI feature surfaces its output as a proposal — review before accepting. Larger models cost more per call; smaller ones (gpt-4o-mini, gemini flash, llama 8b) are usually fine for cleanup and scoring, while frontier models earn their keep on backlog generation and scan repository runs.

Integrations

GitHub integration

Link a repo, then let VibeAssist detect completed work straight from your commits.

Connect a repository to a project to unlock automatic task tracking. From a project, open the GitHub panel, authenticate via OAuth, and pick a repo.

Scan Repository

Click Scan Repository and VibeAssist will fetch recent commits from your default branch, compare them against open tasks, and propose which ones look done. Approve to move the task to Completed (the AI's justification is appended to its notes). Reject to leave it untouched.

Default branch only

GitHub issues created by the workflow only auto-close when commits land on the repo's default branch (usually main) and the message contains Resolves #<number>.

Integrations

GitHub webhook (AI ghost sync)

Let GitHub tell VibeAssist when code lands or issues close, so your backlog updates itself.

The GitHub webhook is the bridge that keeps your backlog in sync with what's actually happening in your repo — no manual copy-paste, no stale statuses. When GitHub fires an event (a push, an issue closing, a comment), it pings VibeAssist, which verifies the request and updates the right tasks automatically.

What it does

  • Reacts to commits on your default branch. Pulls the diff via GitHub's compare API, runs AI analysis on the changed code, updates the backlog, and refreshes the “Synchronised” badge.
  • Closes the loop on AI-ready issues. When an issue tagged vibeassist-ready is closed, VibeAssist finds the matching implementation batch, extracts its TECH NOTES, appends them to the batch, and marks tasks as pending review or completed.
  • Captures issue comments. New comments on vibeassist-ready issues are appended to the batch notes and queued for review.
  • Auto-installs on request. Once the right token scopes are granted, VibeAssist can install or update the webhook on GitHub for you — no manual copy-paste required.
  • Health check. A live status badge shows the last delivery time and event, so you know the connection is healthy at a glance.

Token permissions required

To let VibeAssist list repos, push issues, and manage the inbound webhook, your Personal Access Token needs one of these permission sets:

  • Fine-grained PAT. Contents: Read & Write, Issues: Read & Write, and Webhooks: Read & Write for the repository.
  • Classic PAT. repo plus admin:repo_hook.

Without Webhooks: Read & Write the auto-installer will fail

You can still add the webhook by hand from GitHub, but the one-click install/update button requires the webhook scope. If you see a 403, re-issue the PAT with the missing scope.

Use the production webhook URL

Preview URLs are protected and GitHub deliveries to them can fail with a 302 redirect. The GitHub dialog now shows the stable production Payload URL, even when you're using the preview editor.

Setting it up

  1. 1

    Open project settings

    Go to your project, then Settings → GitHub.

  2. 2

    Save a PAT with webhook scope

    Create the token as described above, paste it into VibeAssist, and save.

  3. 3

    Link a repository

    Select the repo VibeAssist should watch.

  4. 4

    Install the webhook automatically

    Click Install / update webhook automatically. VibeAssist will create or update the webhook with the correct URL, secret, and events.

  5. 5

    Or add it manually on GitHub

    If you prefer, go to Settings → Webhooks → Add webhook in the repo. Paste the Payload URL shown in VibeAssist, set Content type to application/json, and select the Push, Issues, and Issue comments events.

  6. 6

    Verify

    GitHub sends a ping and the health badge flips to Live. Push a commit or close an issue to see activity roll in.

Default branch matters

Push-driven sync only runs on commits to the repo's default branch (usually main). Work on feature branches is picked up after it's merged.

Security

Every request is verified with an HMAC signature (x-hub-signature-256) against your project's stored secret — even though the endpoint is public, only events signed with your secret are processed.

What happens when an issue closes

When an issue with the vibeassist-ready label is closed (typically by a coding agent that finished its batch), VibeAssist:

  • Finds the implementation batch. Matches the issue to the batch it represents.
  • Extracts TECH NOTES. Parses the agent's structured completion notes.
  • Appends them to the batch. So you have a permanent record of what shipped.
  • Updates task statuses. Tasks are marked pending review, then completed by the AI parser if confidence is high.

If anything is ambiguous, the batch stays in pending review so you can confirm before it lands.

Core workflow

Knowledge base

Persistent context — architecture, personas, brand, decisions — that keeps the AI grounded.

The Knowledge Base is where you keep the durable context the AI needs every time it generates or critiques tasks. Add documents manually, or pull them in from your linked GitHub repo.

Add a document manually

  1. 1

    Open Knowledge Base

    From your project sidebar.

  2. 2

    New document

    Title, category, content. Markdown supported.

  3. 3

    Save

    It's instantly available to every AI feature.

Import from GitHub

If your repo has READMEs or a /docs folder, click Import from GitHub, tick the files you want, and import. Each file becomes its own knowledge document using the filename as the title.

Generate a docs prompt

Click Generate Docs Prompt and VibeAssist reads your epics and tasks to assemble a ready-to-use prompt. Copy it into ChatGPT, Claude, or Gemini to draft user-facing docs from what you've actually shipped.

Core workflow

Coding sessions

Plan a focused work block, log the prompts and lessons as you go, and export the result as markdown.

Sessions are short, scoped work blocks attached to a project. Open Sessions in the sidebar to start one. Each session captures its brief, the prompts you sent to your coding tool, the diffs or notes that came back, and the lessons you want to remember — all in one place so you can review or export it later.

How sessions feed the AI

The brief you fill in (goal, current task, technical constraints) is stitched into every prompt the session generates, so the model stays on-rails instead of inventing its own scope. Lessons learned written at the end roll forward into the project's context briefs, so the next session starts smarter than the last one.

Typical flow

  1. 1

    Start a session

    From the project sidebar, click Sessions and create a new one. The form requires goal, task and constraints — that's the brief the AI will reuse.

  2. 2

    Log as you work

    Paste prompts, snippets and outcomes into the session as you bounce between VibeAssist and your coding tool.

  3. 3

    Capture lessons

    When you finish, write down what worked, what didn't, and any follow-ups. These feed future context briefs.

  4. 4

    Export to markdown

    Click Export markdown on the session page to download a single .md file containing the brief, prompts, outcomes and lessons.

What to do with the exported markdown

  • Hand off to your coding tool. Paste the .md into Lovable / Cursor / Claude chat as continuation context for the next iteration.
  • Attach to a PR or retro. Drop it into the PR description or your team retro doc — it doubles as a decision log.
  • Import into the Knowledge Base. Upload it as a knowledge document so future sessions inherit the lessons automatically.
  • Archive for audit. Keep it next to the commit; the brief + prompts make AI-assisted changes auditable later.

One session per shipped batch

Keeping sessions narrow (one feature, one bug, one refactor) makes the export far more useful when you come back to it.

Core workflow

Project configuration

Vibe Templates, architectural constraints, and the design patterns and libraries that drive the Cheat Sheet.

Open Configuration from the sidebar to record decisions that should follow this project everywhere. These values are injected into AI prompts and rendered on the Cheat Sheet.

What you can set

  • Vibe templates. Per-tool instructions for Lovable, Cursor and Other tools — appended to generated prompts.
  • Architectural constraints. Non-negotiable rules the AI must respect (one per line).
  • Design patterns. Conventions humans and the AI should follow (rendered as a list on the Cheat Sheet).
  • Chosen libraries. Default building blocks for the project (rendered as chips on the Cheat Sheet).

Treat it as living documentation

Update Configuration whenever the project takes a new direction — the AI's suggestions only stay sharp if its constraints stay current.

Core workflow

Project cheat sheet

A read-only, at-a-glance reference of the design patterns, libraries and constraints chosen for this project.

The Cheat Sheet renders whatever you've recorded in Configuration into a clean, scannable page. Use it as the one-stop reference when you (or a teammate) are about to write a prompt, open a PR, or onboard onto the project.

What it shows

  • Chosen libraries. Rendered as colored chips so the default toolkit is obvious.
  • Design patterns. Bulleted list of conventions to follow.
  • Architectural constraints. Bulleted list of non-negotiable rules.
  • Vibe templates. The exact per-tool instructions appended to generated prompts.

Click Edit in the top right to jump back into the Configuration screen. Nothing on the Cheat Sheet is editable in place — it's intentionally read-only so it can be linked and shared without fear of accidental changes.

AI & automation

UI consistency review

Scan the linked repo for UI pattern deviations and approve AI-proposed fixes.

UI Review scans the connected codebase for visual and component-level inconsistencies — a button using a raw color where a token should be used, a dialog with the wrong spacing scale, a section with hardcoded dark-mode classes. The AI proposes specific fixes; you approve or reject each one.

Uses your AI provider

Scans call your provider and consume tokens. Run them after notable UI changes rather than on every push.

AI & automation

AI usage

Per-project breakdown of token consumption and estimated cost across every AI model you've used.

Open AI usage from the Workspace group in the sidebar to see how tokens have been spent on this project. The table groups by model and feature (task generation, repo scan, deep ingest, UI review, embeddings, …) and shows running totals so you can spot the expensive ones early.

Estimates, not invoices

Costs are computed from public per-token pricing for each model. Your provider's actual bill is the source of truth.

Help

Tips for power users

Small habits that compound: keep context fresh, scan often, trust your reviews.

  • Keep your knowledge base current. Richer context = smarter AI.
  • Scan repos regularly. Fastest way to keep statuses honest.
  • Review every proposal. They're suggestions, not commands.
  • Mobile triage. The layout is built for between-meeting cleanup.

Keyboard shortcuts

?Open keyboard shortcuts
/Focus search
NNew task in the current backlog
GthenDGo to dashboard
GthenSGo to settings
EscClose dialogs
Shortcuts work anywhere except inside a text field.
Help

Troubleshooting & FAQ

The most common gotchas and how to clear them in under a minute.

GitHub says the last delivery failed with Invalid HTTP Response: 302.
The webhook is pointing at a preview URL. Open Settings → GitHub and click Install / update webhook automatically, or manually replace the Payload URL with the production URL shown in the dialog.
The webhook auto-install button shows a 403 error.
Your PAT is missing the webhook scope. Re-issue it with Webhooks: Read & Write (fine-grained) or admin:repo_hook (classic), then save the new token in VibeAssist and try the auto-install again. You can still add the webhook manually on GitHub using the URL and secret shown in the dialog.
My GitHub issue didn't auto-close.
Check that the commit was merged into the default branch and that the message contains Resolves #<number>.
The AI is suggesting duplicate tasks.
Tighten up your existing task titles — the AI uses them to detect overlap. Vague titles produce vague matching.
A repo scan didn't find a task I expected.
Approve/reject what came back, update the missing ones manually, and re-scan. Subsequent scans pick up newer commits.
Can I use a different AI provider?
OpenAI, Anthropic (Claude), Google (Gemini), OpenRouter, xAI (Groq), DeepSeek, Mistral and Perplexity today. More providers are on the way.
Is my data isolated?
Yes. Every project is scoped to its owner with row-level security. Nobody else can see your backlog, knowledge base, or keys.
Still stuck? Open Settings to manage your account, or reach out from the help menu. Happy shipping. 🚀