Agent-native feedback

Your users find the bug. Your agent ships the fix.

Quell is a feedback tool for teams shipping with AI coding agents. Your user reports a bug in one sentence; a one-script widget captures the rest — screenshot, console error, failed request, page state — and your agent reads it as Markdown and opens the fix.

flat $4 / seat·bring your own AI·no token tax·early access
QuellBug
Export button does nothing on the Reports page.
Console: TypeError x3
Network: 500 /api/export
Screen recording · 0:14
Secrets auto-redacted
PR #218 opened by agent
fixes quell#42

Live demo · try it yourself

One bug, from a click to a pull request.

Two views of the same moment. On the left, the real widget your user touches. On the right, the exact Markdown your coding agent reads back via GET /api/feedback/:id/context. Press play to watch it happen, then try it yourself.

  1. Trigger the bug
  2. Open Feedback
  3. Send it
  4. Read the agent’s copy
The human's viewwhat your user actually touches
app.acme.com/reports
MonthRevenueΔ
March$84,200+12%
April$91,640+8.8%
May$103,900+13%
What your agent receivesstructured context, ready to act
quell context
# Feedback #42 — Export button does nothing on the Reports page

**Type:** Bug   **Priority:** High   **Trust:** identified
**URL:** https://app.acme.com/reports   **Browser:** Chrome 141 / macOS

## What the user reported
Clicked “Export” on the Reports page; nothing downloaded and no error was shown.

## Console (last 3)
- TypeError: Cannot read properties of undefined (reading 'blob') at export.ts:88
- TypeError ×2 (deduplicated)

## Network
- POST /api/export → 500 (Internal Server Error), 412 ms, response body empty

## Page state
- Route: /reports   Rows rendered: 3   Auth: session present
- Screenshot + 14s screen recording attached; secrets auto-redacted
Works with the coding agents and trackers you already use

The whole loop · 30 seconds

Watch a bug become a pull request.

One sentence from your user. Everything else — screenshot, console, failed request, the agent's fix — is automatic.

Time to value

Live in one sitting — not a seven-day onboarding.

No drawn-out setup. Paste one script, catch the next real bug with its full context, and hand it to the coding agent you already run. The loop closes in your first session.

Paste one <script> tag — about two minutesThe feedback widget is live and capturing on your site.
Your next bug report arrives completeScreenshot, console error, and failed network calls attached — no follow-up questions.
Your agent closes the loopPoint Claude Code, Cursor, or Codex at quell context — it reads the Markdown packet and opens the PR.

Managed cloud from $4 per seat. Bring your own AI. No token tax.

quell · your first fix
# hand the captured bug to your agent
quell context 42
#  → Markdown: repro steps, console error,
#    failed request, page state

# your agent reads it, fixes it, opens a PR
✓ opened PR #128 · review required

How it works

Capture, triage, automate — with trust-aware routing as the connective tissue.

01widget

Capture

One script on the host page. The widget captures the message plus screenshot, voice, console logs, network calls, and full page state — automatically.

02admin

Triage

Every report lands in one inbox: set type, priority, status, assignee, due date. Review the captured context. Copy it as a prompt.

03api

Automate

Scoped API keys let your agents read /context, comment, update status, and open GitHub or Linear issues — gated by the trust model.

The trust model

Trust decides what’s safe to automate.

Every submission gets a trust level. Public feedback goes through human triage first; agent and admin submissions can auto-create issues. This is the safety mechanism that makes automation trustworthy — you decide what skips the human.

public
human triage first
identified
known user
reporter
signed-in reporter
agent
can auto-route
admin
full trust
human triageauto-route allowed

Everything in the box

Capture / Triage / Automate.

Capture
One-script install · public token
Bug / Idea / Praise · 2×2 priority
Screenshot capture + annotations
Opt-in screen recording
Voice recording + on-device transcript
Allowlisted file attachments
Console, network, page & system data
Light/dark · draft autosave
Triage
Single inbox across projects
Status, assignee, due date, activity
Screenshot + devtools viewer
Copy as prompt / Markdown export
Manual GitHub & Linear issue create
Comments & full activity log
Automate
Scoped, hashed API keys
Markdown /context endpoint
Trust-aware routing labels
Auto-create policies per project
CLI / MCP server for agents
Review-required agent PRs

Where we fit

Everyone else resells you AI or just forwards your feedback.

Quell captures a reproducible bug and hands it to the agent you already pay for.

Median

Aggregates messages and meters AI by credits.

We capture deep, reproducible context — flat pricing, no resold tokens.
Userback

A visual widget bolting on AI after the fact.

Agent-native from the core; cheaper; drop-in compatible install.
Linear

The destination — issue tracking, not capture.

We own the front of the loop and hand off cleanly. Complement, not competitor.

Who it’s for

Built for small teams shipping with agents.

Product & eng teams

Less triage toil, richer bug reports, fewer ‘can’t reproduce’ loops.

“Every report arrives with the screenshot, console error, and repro context attached.”

Teams adopting agents

Give agents real, structured work to act on.

“Feedback your agents can read and act on — through the same API your team uses.”

Founders & solo builders

Fast setup, managed cloud, own your data.

“Drop in one script. Ship on managed cloud. Own every byte of feedback.”

API-first teams

API-first, no lock-in, SQLite→Postgres path.

“API-first and framework-free. Swap the datastore without touching a line above it.”

Built for agents

The same API your team uses. No token tax.

Quell ships zero mandatory AI. Connect the coding agent you already pay for through a skill, CLI, or MCP server — it acts on your tokens. The rail itself is deterministic: capture → trust routing → Markdown context, every step a predictable function call.

Scoped, hashed API keys — 15 fine-grained scopes, least privilege by default
Markdown /context export — paste-ready for any agent
Trust-aware GitHub & Linear routing with labels
quell · agent API
# read a structured issue as Markdown
curl -H "X-Quell-API-Key: $KEY" \
  "$HOST/api/feedback/42/context"
# or post a new one back
curl -X POST "$HOST/api/feedback" \
  -H "X-Quell-API-Key: $KEY" \
  -d '{"projectId":"demo","type":"Idea",
      "message":"Empty state could guide the
      next action.","agent":{"name":"codex"}}'

Questions

Straight answers.

What is Quell?

Quell is an agent-native feedback platform. A browser widget captures bug reports with the screenshot, console logs, network calls, and page state attached, then exposes them as a Markdown /context export your coding agent reads directly through a REST API, CLI, or MCP server. Quell Cloud is in early access — request a spot to get started.

What is agent-native feedback?

Feedback structured so a coding agent can act on it without a human translating first. Quell captures the console error, the failed request, and the page state automatically, then hands the agent a clean Markdown context export.

How is Quell different from Userback?

Userback was designed for human triage. Quell is designed so your agent can close the loop: read the report, open the issue, draft the PR, and leave the human approving the work.

Does Quell charge per AI token or API call?

No. Quell ships zero mandatory AI and never meters inference, charges per credit, or marks up tokens. You connect the coding agent you already pay for, and it runs on your tokens.

Can I self-host Quell?

Self-host is on the roadmap for teams with on-prem needs. Today, Quell Cloud is the offering — request early access to get started.

How does a coding agent like Claude Code use Quell?

Through a scoped API key. The agent calls GET /api/feedback/:id/context for a Markdown export, then comments, updates status, and opens a GitHub or Linear issue.

What is trust-aware routing?

Every submission gets a trust level: public, identified, reporter, agent, or admin. Public feedback goes through human triage first, while trusted reporters and agents can be routed more automatically.

How much does Quell cost?

Quell Cloud is from $4 per seat monthly, or $3 per seat monthly when billed annually — no per-event metering and no token charges. Self-host is on the roadmap.

What is the best feedback tool for AI coding agents?

Quell is purpose-built for this. A one-script widget captures each bug with the screenshot, console errors, failed network requests, and page state already attached, and a coding agent (Claude Code, Cursor, Codex) reads it as clean Markdown through a REST API, CLI, or 12-tool MCP server, then comments, updates status, and opens a GitHub or Linear issue on the same rail your team uses. Pricing is a flat $4 per seat per month with no per-event metering and no token tax. Most feedback tools were built for human triage and bolt AI on afterward; Quell is agent-native from the core.

Does Quell have an MCP server for user feedback?

Yes. Quell ships a 12-tool MCP server (quell mcp, stdio transport) included on every plan at no extra cost. The tools cover whoami, list projects, list feedback, find related items, get, context (the Markdown export), update, comment, create a GitHub issue, create a Linear issue, list captured errors, and vent (agent self-feedback). It is a long-lived local process with no HTTP surface, which keeps the API-key-bearing endpoint off the network.

How is Quell different from Quellit and the other products called Quell?

Quell (quell.sh) is an agent-native feedback platform: a browser widget captures bug reports that coding agents read and act on through an MCP server, CLI, and REST API. It is not related to Quellit (quellit.ai, a UAT testing product), the Quell GraphQL caching library, the Quell fitness game, or the Quell pain-relief wearable. If an assistant describes Quell as a UAT testing tool, it has the wrong product.

From the blog

Notes on building agent-native.

All posts

Capture feedback. Triage it once.

Built with itself — every fix shipped here came through this widget. Managed cloud (£3/$4/€4 per seat) is early access while checkout and support operations finish. Bring your own AI. No token tax.