🏗️
Implements
Builder
Checks out an isolated worktree, writes the code, and opens the PR.
Autonomous engineering throughput, governed and observable
Autonomous engineering throughput, governed and observable.
Quester dispatches a crew of specialized AI agents from inside Linear — planning, building, reviewing, and QAing your code — with per-task USD cost tracking and self-hosted execution so spend and security are never a black box.
Linear-native dispatch
Self-hosted execution
Per-task cost tracking
Bring your own model
Human-authorized merges
MUL-241Add dark mode toggle
Linear issue@Builderdispatched•effort: M•model: claude-sonnet
✓ Planning2s
✓ Coding41s
✓ PR opened: #182 "Add dark mode toggle"
✓ Reviewer: 2 comments · approved
✓ QA: passed
4m 31s total$0.14
7
specialized agents
4
effort tiers (S/M/L/XL)
$0.00
per-task cost tracking
For engineering leaders
AI agents are powerful. Ungoverned, they're a liability.
Most AI coding tools are glorified autocomplete — hard to observe, impossible to cost, and running on infrastructure you don't control. Quester is different: a real pipeline dispatched where work already lives, with audit trails, cost accounting, merge gates, and self-hosted execution.
Visibility
Without Governance
Agents run in black boxes
With Quester
Every session streams activities live into Linear
Cost Accountability
Without Governance
No cost visibility
With Quester
Per-run USD cost computed from token usage
Code Custody
Without Governance
Code leaves your infra
With Quester
Agents run on workstations you own
The pipeline
From @mention to merged PR in four steps.
1
Mention or assign
@mention a bot on the Linear issue. Quester auto-classifies effort as S/M/L/XL and routes to the right model tier.
2
Dispatch
The orchestrator picks up the job, checks out an isolated worktree, and starts the agent session. Thought, action, and response stream live into Linear.
3
Build, review, test
Builder implements and opens a PR. Reviewer posts structured line-level comments. QA checks out the branch and runs your verify command. Fix loops run up to 4 rounds before escalating to a human.
4
Authorize & ship
React 👍 on the Linear issue comment to authorize the squash-merge. The PR lands, the issue closes, and cost is logged.
1
Mention or assign
@mention a bot on the Linear issue. Quester auto-classifies effort as S/M/L/XL and routes to the right model tier.
2
Dispatch
The orchestrator picks up the job, checks out an isolated worktree, and starts the agent session. Thought, action, and response stream live into Linear.
3
Build, review, test
Builder implements and opens a PR. Reviewer posts structured line-level comments. QA checks out the branch and runs your verify command. Fix loops run up to 4 rounds before escalating to a human.
4
Authorize & ship
React 👍 on the Linear issue comment to authorize the squash-merge. The PR lands, the issue closes, and cost is logged.
Quester agents
Seven specialists. One coordinated pipeline.
Each agent has a defined role, dispatched in sequence and able to hand off to the next. Rename them, replace them, or extend the crew to match how your team ships.
🔍
Reviews
Reviewer
Posts structured line-level PR comments with severity labels. Approves or requests changes.
🧪
Validates
QA
Checks out the branch, runs your configured verify command, and reports pass/fail.
📋
Plans
Planner
Explores the codebase, writes an implementation plan, and estimates effort.
🔬
Diagnoses
Investigator
Diagnoses failures, audits incidents, and proposes follow-up sub-issues.
📡
Researches
Researcher
Surveys external APIs, libraries, and technologies to surface options before implementation.
⚔️
Stress-tests
Dissenter
Adversarially reviews a plan before code is written — argues the other side so weak assumptions surface early.
The roster is configurable. Keep these personas, rename them, or wire your own specialist agents to match how your team ships.
Feature highlights
Built for teams that need more than autocomplete.
Quester is an observable, cost-accounted, self-hosted pipeline — not a chat interface.
Native Linear integration
Agents are real Linear users. @mention dispatch, native agent sessions, session plans, follow-up sub-issues, and emoji-driven merge authorization — all inside Linear.
Automatic fix loops
Builder → Reviewer → QA → fix. Up to 4 rounds automatically before escalating to a human. The feedback loop runs without you.
Effort-based model routing
Small tasks use faster, cheaper models. Large tasks get top-tier reasoning. Quester classifies effort (S/M/L/XL) and routes accordingly.
Per-task cost transparency
Every run's real USD cost — computed from token usage — is shown per issue, per agent, and per repo. Spend is never a black box.
Bring your own model
Claude, OpenAI/Codex, Google Gemini, Perplexity, or any LiteLLM-compatible endpoint. One adapter layer, no lock-in.
Self-hosted, multi-machine
Central orchestrator plus workstation nodes you own. Code, credentials, and transcripts never leave your infrastructure.
Durable & zero-downtime
Webhook queue survives outages without losing events. Hot code reload keeps running agents alive through server updates.
Live dashboard
Active runs, history, cost per issue, per-repo config, provider settings — all real-time via SSE. Know exactly what every agent is doing.
Governance
Humans stay in the loop where it matters.
Quester automates the repetitive parts of engineering — not the decisions. Every merge requires explicit human authorization. Every run is logged. Code and credentials stay on your infra.
Human-authorized merges
React 👍 on the issue to authorize a squash-merge. Agents cannot merge without explicit approval.
Isolated worktrees per run
Every agent session runs in a fresh git worktree. No shared state between runs.
Self-hosted custody
Code, credentials, and agent transcripts stay on infrastructure you control. Nothing leaves your network.
Full audit trail in Linear
Every session, handoff, and decision is logged as a Linear comment. The full history is always readable.
Integrations
Built for your stack
AGENTS.md & CLAUDE.md
Agents read your repository-level instructions before writing any code.
Per-repo "zone" config
Configure branch names, verify commands, and per-agent enablement.
Managed GitHub identities
Keep commits clearly attributed using dedicated per-agent GitHub profiles.
Documentation
Explore the docs.
Everything you need to deploy, configure, and extend Quester.
Getting Started
Install Quester and ship your first issue in minutes. Setup guide, config, and first run.
Architecture
How the server, agents, dashboard, and worker fit together. Data flow and component map.
Agent Reference
What each bot does — peon writes code, grunt reviews, headhunter runs QA, shaman plans.
Configuration
Config files, secrets, OAuth tokens, and environment variables explained.
Dashboard Guide
Monitor agents, active quests, run history, and workstation health in real time.
Open source
The full agentic engineering stack, open source.
Quester is a real system with real moving parts: orchestrator, web console, CLI, worker, and workstation execution. The repository shows how the pieces fit together.
Repository
simontzky/quester
The Linear-native system that turns issues into reviewed pull requests through configurable agent orchestration.
0 stars
0 forks
Stack map
What ships inside the repo
multi-runtime
Elixir server
Phoenix orchestration, webhooks, sessions, follow-up chains, and dashboard API.
Web console
A focused operator surface for reviewing runs, personas, and workstation state.
CLI
A direct operator surface for booting the server, workstations, and local workflows.
Cloudflare Worker
Durable webhook buffering so events survive internet or tunnel instability.
Early access
Request access to Quester.
Quester is in active development. Engineering teams can request early access to self-host and run the full agent pipeline on their infrastructure.
Phase 01
Self-hosted private beta
Run the full agent pipeline — orchestrator, workstations, all 7 agents — on infrastructure you own.
Phase 02
Quester Cloud
Shared orchestration, team dashboards, and managed infrastructure for teams that want to skip the setup.
Follow the build