QUESTPIE
Star on GitHubStart free
03 · Autopilot · Early access

Run your company.Build your apps.Automate the rest.

Autopilot is the AI-native operating layer for QUESTPIE. Think Linear for work, Lovable for app building, Zapier for automation. On one open source stack. Self-host, or run hosted on Cloud.

Like
Linear + Lovable + Zapier
License
MIT
Audit
Every step
QAP-4821Publish Q3 launch announcementIn review
Type
Feature
Project
Marketing site
Workflow
review-content
Updated
2m ago
Activity
08:12MarekCreated issue · attached workflow review-content
08:12step.rundraft ✓ generated 612 words
08:14step.runfact-check ✓ 3 sources verified
08:15step.runseo-pass ✓ title + meta updated
08:16step.waitForEventhuman-approval
·step.runpublish · queued
trace: wf_a821-0428 · 5 / 6 steps5.4s
QUESTPIE is open source. Star us on GitHub.
The fastest way to support a small team building everything in public.
questpie/questpie
The product

Five surfaces. Plus everything underneath.

Top-level navigation is what a human actually thinks about: work, procedures, triggers, context, scopes. Agents, builder, packs and integrations live inside the issues that need them.

Issues

The work, in one keyboard-driven list.

Dense rows. Status icons that scan in a glance. Quick filters for Open, Needs review, Needs attention, Done, Scheduled. Open any issue to see properties, activity, attached workflow runs.

/admin/issues
OpenNeeds reviewNeeds attentionDoneScheduled
search · ⌘K
IDTitleStatusProjectWorkflowUpdated
QAP-4821Publish Q3 launch announcementIn reviewMarketing sitereview-content2m
QAP-4818Fix DNS for marekstreet.barbersIn progressCloud opsincident-triage12m
QAP-4815Weekly newsletter draftTodoMarketing sitedraft-newsletter1h
QAP-4801Investigate slow listing queriesBacklogRealtor app·yesterday
QAP-4799Review June P&LDoneInternalmonthly-financeyesterday
QAP-4790Onboard new partner shopIn progressPartnershipspartner-onboarding3h
6 of 42 issues · sort: updated descshortcut: ↑↓ Enter
BacklogTodoIn progressIn reviewDoneCancelled
Workflows

Procedures, written once. Re-attached to every issue.

Each step persists its result. The process can sleep for hours, wait for human approval, invoke sub-workflows, and survive every restart. Compose business logic the same way you write any TypeScript.

workflows/review-content.ts
1// workflows/review-content.ts. Durable, restartable
2export default workflow(
3 "review-content",
4 async (step, { issueId }) => {
5 const issue = await step.run(
6 "load", () => ctx.collections.tasks.findById(issueId)
7 );
8
9 const draft = await step.run(
10 "draft", () => agent.draft(issue, { capability: "writer-v2" })
11 );
12 await step.run("fact-check", () => agent.factCheck(draft));
13 await step.run("seo-pass", () => agent.seo(draft));
14
15 await step.waitForEvent("human-approval");
16 await step.run("publish", () => ctx.collections.tasks.update({ id: issueId, status: "done" }));
17 }
18);
step.run(name, fn)Execute & persist. Safe to retry.
step.sleep(name, '14d')Pause hours, days, weeks.
step.waitForEvent('approval')Block until a signal arrives.
step.invoke('sub-flow', input)Compose workflows.
ctx.collections.*Read & write any QUESTPIE collection.
agent.run(capability, input)Hand off to a configured capability.
Workflows live in workflow_configs as steps + defaults. Attach one to any issue, or trigger via Schedules.
Schedules

Time triggers that produce real work.

A schedule either creates an issue with a template, or runs a workflow directly. Pick cron, pick a workflow, decide what happens on overlap. The product surface stays human.

NameCronActionWorkflowNext run
Weekly newsletter brief0 8 * * 1create issuedraft-newsletterMon 08:00
Monthly P&L review0 7 1 * *create issuemonthly-financeJul 1, 07:00
SEO sweep0 4 * * *run workflowseo-sweeptomorrow 04:00
Backup verification0 */6 * * *run workflowverify-backupin 4h

Knowledge

Context, docs, brand rules, references. Indexed for semantic search. Workflows and agents query knowledge before they decide.

knowledge.search("how do we structure pricing pages") →
3 results · pricing-style.md · landing-copy-v2.md · brand-voice.md

Projects

Workspaces, repos, scopes. Each project has its own issues, workflows, schedules and defaults. Multi-tenant by design.

Marketing sitedefault8 open
Cloud ops14 open
Realtor app23 open
Partnerships4 open
AI builder

Like Lovable, but it edits real QUESTPIE code.

Describe a change and the builder modifies collections, admin views, routes, components and content as a real commit. Live preview, human approval, full audit. Hosted on Cloud or self-hosted.

builder · marekstreet.barbers24 edits
Tighten the hero, add a contact form, swap headings to Geist.
Updated hero spacing, inserted a contact form below the pricing, applied Geist to h1, h2 and h3. Live preview ready.
+ Geist h1, h2, h3+ <ContactForm/>~ hero padding 80 → 120
Add a Team section with three barbers and short bios.
Added a Team section with three columns and image placeholders. Hooked to the `team` collection so you can edit in admin.
+ <TeamSection/>+ collections/team.ts (3 rows)+ image slots
Describe the change…
3 changes pendingDiscardApply & deploy
MAREK STREET BARBERSHOME · BOOK · MENU · ABOUT
Old-school cuts.New-school booking.

Three chairs. Twenty years. Walk in or reserve online in seconds.

Book a chairSee gallery
+ just addedThe team
Marek
PHOTO-1
Tomáš
PHOTO-2
Bea
PHOTO-3
● live preview · marekstreet.barbersp95: 84ms · 0 errors
Add an opening-hours widgetTranslate to SlovakUse a warmer paletteAdd Instagram embed
Integrations

Where your team already talks. Where your code already lives.

Autopilot connects to the channels your humans use and the systems your code runs in. Issues, approvals, digests, deploys. Flow through Slack, GitHub, email and MCP.

SlackBeta

Open issues from a thread. Approve workflow steps with one click. Get digests on a channel.

GitHubBeta

Pull requests created by the builder. Issues sync to QUESTPIE. Branch-per-issue.

GitLabPlanned

Same surface as GitHub for self-hosted GitLab installations.

LinearPlanned

Two-way sync for teams already living in Linear. Move when ready.

EmailActive

Inbound mail creates issues. Outbound from workflows uses configured sender.

MCPActive

Your data exposed as tools to external IDE-side agents. Same RBAC as humans.

Packs

Business capability, installable in one command.

A pack bundles skills, workflows, schedules, MCP tools and knowledge seeds into a single installable unit. Drop one in, get a working capability. Share with your team or the community.

@questpie/pack-marketing-contentComing soon

Editorial calendar workflow, SEO sweep, brand voice prompts, newsletter schedule.

3 workflows2 schedules1 agent profileknowledge seed
@questpie/pack-support-opsComing soon

Inbound triage from Slack/email. Knowledge-grounded answers. Escalation rules.

4 workflows1 schedule2 MCP tools
@questpie/pack-release-engineeringComing soon

Generate changelogs, drive deploys via Cloud, post digests to Slack.

5 workflowsCloud bridgeagent: release-manager
@questpie/pack-finance-monthlyComing soon

Pull P&L, draft monthly review, send for approval, archive in knowledge.

1 workflow1 schedule (cron 0 7 1 * *)approvers config
$ autopilot pack install @questpie/pack-marketing-contentregistry · @questpie / public · private soon
Agents · Runtime · MCP

Agents your data trusts. Models you choose.

An agent is a profile: capability, allowed tools, MCP scope. Point it at a provider. Anthropic, OpenAI, local. And let it work through the same RBAC rules that apply to your humans.

Providers
Bring the model. Or run it on-prem.
providers · capabilities · models
Anthropic
Claude SDK
Active
OpenAI
Codex SDK
Active
Local
Ollama et al.
Active
Custom
BYO endpoint
Active
Gemini
Vertex / API
Planned
Cursor
Agent SDK
Planned
MCP
Your data, safely exposed.
model context protocol
collections:
tasks: read, write
knowledge: read, write, delete
schedules: read, write
projects, runs, providers: read
routes:
exposeAnnotated: true
resources: schemas + routes

Same RBAC as humans. Annotated routes auto-discovered as tools. Schemas and OpenAPI shipped as MCP resources, so agents know your data model without you wiring prompts.

Developer cockpit · planned

A terminal cockpit, in the same package.

The same product, operated from your terminal. Open in any repo, inspect runs, steer sessions, retry failed steps. Keyboard-first, session-scoped. Ships with the Autopilot package.

~/dev/marekstreet · autopilotmock
■ QUESTPIE AUTOPILOT
marekstreet / marketing-site / session 676dcda1 / anthropic

State
done 4 ready 15 failed 1 todo 40

Current
task: QAP-246
phase: validation
last event: Bash

Ready
QAP-256 draft weekly newsletter
QAP-247 bug DNS edge-case
QAP-248 task onboard partner shop

Failed
QAP-238 Primary validation failed

$ /run-task QAP-256
$ autopilotopen the TUI in the current workspace
$ autopilot statushealth, current run, ready tasks
$ autopilot run-task QAP-246execute one issue in a fresh session
$ autopilot logs --followcondensed session log, conversation mode
$ autopilot retry QAP-238re-run a failed task with remediation
$ autopilot note QAP-246 'focus only on HookContext'persistent steering for the next step
Same product mental model (Issues, Workflows, Schedules), session-scoped logs, command palette with autocomplete on issue IDs and projects. Coming with the Autopilot package.
Open source · Two paths

Self-host the open package. Or run hosted on Cloud.

Autopilot is MIT. Self-host as long as you like. Workflows are TypeScript, schemas are Drizzle, everything stays in git. When you outgrow one machine, point Cloud at the same project. Hosted Autopilot inherits all of it: same code, same data, plus managed runtime and the Cloud deploy bridge.

# Self-host the open package
$ bun add @questpie/autopilot
$ bunx questpie migrate:up
$ bun dev
# Open the TUI cockpit (coming with the package)
$ bunx @questpie/autopilot
FAQ

Questions, answered.

Honest answers. No marketing fluff. Can't find what you need? Ask the team.

An AI-native operating layer for QUESTPIE. Like Linear for work, Lovable for app building, Zapier for automation, on one open source stack. Issues are the human entry point, workflows are durable procedures, agents are configurable profiles, and the AI builder edits your real code.

Issues live in a 'tasks' collection (title, description, type, status, project, workflow). Workflows in 'workflow_configs' (steps + defaults). Schedules in 'schedules' (cron + action). Knowledge in 'knowledge'. Projects in 'projects'. Plus capabilities, providers, models, runs and events under advanced settings. Everything is a normal QUESTPIE collection.

Similar vibe, different surface: the builder reads your QUESTPIE repo and edits collections, admin config, components and routes. Every change is a workflow run with named steps, a preview deploy through Cloud, and an approval before publishing.

An installable bundle of capability: skills, workflows, schedules, MCP tools, knowledge seeds and recipes. Drop a pack in and you get a working capability. Marketing-content, support-ops, release-engineering, finance-monthly. Packs are shareable across the team or the community.

Slack and GitHub are first-class. Open issues from a thread, approve workflow steps with a click, pull requests created by the builder. Email, Linear and GitLab are on the roadmap. MCP exposes your data to external IDE-side agents with the same RBAC.

Because real ops processes survive restarts, wait for human approvals, sleep for days, and need to be auditable. step.run persists every result. Re-runs skip completed steps. Sleeps and waits don't burn worker time.

Both are first-class. Self-host the open package on your own infra. You wire workers, providers, MCP, packs, schedules yourself. Or run on Cloud and get a managed Autopilot installation per account, with the Cloud deploy bridge wired in.

Providers are pluggable: Anthropic and OpenAI are active, plus local providers (Ollama et al.) and custom endpoints. Gemini and Cursor are planned. Configure per capability. Agents pick the provider they need.

Planned, shipping with the Autopilot package. Open it in any repo to see current issue, ready queue, latest run, condensed logs. Same Issues / Workflows / Schedules mental model, keyboard-driven.

Coming Q4 2026

Be early on Autopilot.

We're onboarding teams in batches. Tell us what you'd run on Autopilot and we'll fast-track the closest fits. No spam, ever.

MIT-licensed·$ bun add @questpie/autopilot·View source ↗