TaskHound - the AI Desk for dev teams

Your team's AI agents
are working alone.

Claude Code on one laptop, Codex on another - in silos, invisible to each other, repeating each other's work. TaskHound gives the whole team one shared environment where everyone's agents work side by side: visible, steerable, and compounding.

Start your team - $19.95/user/mo See the live board ↓

Bring your own agents: Claude Code · Codex · Gemini · Grok - your subscriptions, your keys.

Team board 4 live
acme · TaskHound Container
MK Mara Claude Code

api · feature/auth-refresh

editing auth/session.go

25:4448.2k tok

DV Devon Codex

web · fix/cart-totals

running npm test -- cart

12:4319.7k tok

PR Priya Gemini

api · perf/query-cache

profiling GET /orders (p95 840ms)

35:0161.4k tok

TW Tom Grok

infra · chore/tf-modules

running terraform plan

07:118.9k tok

SL Sam Claude Code done · 3 learnings distilled

web · feat/onboarding-wizard · finished 14 min ago

Co-operative Intelligence

The Co-Working Loop

TaskHound is more than a runner; it’s an interactive AI desk. Developers and agents share context, steer in real-time, and capture learnings so the entire workspace grows smarter together.

Planning Yard · acme-core
2 users online
ME You (Product Owner)
Task: Wire up Stripe Billing

"Please add a webhook-listener at /webhooks/stripe and save a note of Stripe key rules for the team."

MK Mara (Claude Agent)
Planning route tree...

"I've mapped the endpoint routes, identified the Stripe billing library, and will scaffold a test suite."

📋 Specs Analyzed: 100% 🛣 Dependency Map: Created 🤖 Agent Assignment: Ready
Synergy Sandbox

Collaborative Sandbox Command Center

Experience TaskHound's co-working flow. Select a project goal, pick your expert AI agent partner, and operate the human-steering checkpoint yourself to see compounding intelligence in action.

1. Select a Team Goal
2. Assign AI Partner
Collaborative Run Active

Planning Goal Sequence

The agent has initialized. Reading specifications and checking the Team Knowledge Base for historical learnings on this project...

Workspace Monitor · initializing
PLANNING
🐕
Workspace Knowledge Base
Checking indexes...
No distilled learnings loaded yet. Distilled learnings from past runs automatically populate context once verified by humans.
Instant Launchpad

Template-Guided AI Vibe Coding

Skip the boilerplate and dive straight into vibe coding. TaskHound’s smart wizard asks high-level business questions, runs an agent-led deep-dive analysis, and deploys a fully functional runnable slice in seconds.

🌐

Website

Marketing sites, portfolios, or docs. Deploys a live dev server with working nav routes instantly.

Explore Website Template →
🅦

WordPress theme

Bespoke classic or block themes. Pre-installed in local Docker environments, ready to edit.

Explore WP Template →
📱

Web & Mobile App

Custom stateful apps with accounts & database integrations. Runs end-to-end with real data.

Explore App Template →
📊

Dashboard

Internal analytic tools over Postgres/MySQL/Stripe. Renders live charts & filtered data tables.

Explore Dashboard Template →
🕹

2D Game

Platformer, puzzle, top-down or card games. Renders playable vertical slices in Phaser/Godot.

Explore 2D Game Template →
🎮

3D Game & World

Immersive 3D interactive environments. Deploys character controls & generative meshes.

Explore 3D Game Template →
🐧

Linux service

Background daemons, API queues or cron workers in Go/Rust/Python. Under systemd or Docker.

Explore Service Template →
⌨️

CLI tool

Compiled command-line helpers. Preconfigured with help instructions and Dry-run simulation modes.

Explore CLI Template →

Unleash High-Performance Business AI Suites

Integrate custom tools, orchestrate multi-agent collaboration, and let your team design entire apps without manual boilerplate. From small microservices to full game jam vertical slices, this is vibe coding backed by secure enterprise isolation.

🤖 Multi-Agent Orchestration • 🔒 Secure Sandbox Containers • 📊 Auto-Distilling Knowledge Bases
Steering & Authority

The Agent's Toolbox & Human Gate

Empower your agents with powerful local development tools. For sensitive system actions or secret retrievals, TaskHound sits as a human-in-the-loop gate—letting you steer or approve actions in real-time.

File Sandbox · index.js
Code Refactoring
// Agent editing file locally ...
+ import { debounce } from './utils';
function handleResize(e) {
- updateLayoutWidth(e.target.innerWidth);
+ debounce(() => updateLayoutWidth(e.target.innerWidth), 150)();
}
Code formatting matched: Prettier ✓
Linter check passed: ESLint ✓ (0 warnings)

The whole desk, in motion

Watch it work

A live session, the Planning Yard, and the plumbing underneath. Everything below is the real interface, animated.

checkout flow polish Claude 0:00
↑
⬒ payments revamp planning
▲Mara
▲Zack
notes, flows, drawings, chat: one board ✨ Generate Plan
◆ Keystore
STRIPE_KEY project ••••••••

Agents load secrets without ever printing them. Project keys override global ones.

⎇ TaskHound Git
upstream THGit Container

Agents push to THGit, never to your repo. A human-approved push request is the only road upstream.

🕹 One driver
ZP MK Request control
✎

Everyone watches live. One person steers. Handoff takes a click.

The silo problem

Every developer runs AI. Almost no team runs it together.

Agents live on laptops, in personal terminals, behind individual accounts. The result is the same on every team:

Work gets duplicated

Two devs, the same bug, the same prompt, the same afternoon - and neither knows. Every silo re-explains the codebase, re-derives the conventions, re-pays the same tokens for the same answers.

Learnings evaporate

Your agent just figured out the deploy quirk, the flaky test, the auth edge case. Then the session ends, the context dies, and tomorrow's session - yours or anyone's - starts from zero.

Nobody can see it

What is the AI doing on your codebase right now? Which branches did it touch? Who approved what? On laptops, there is no answer - agent work is unreviewable by default.

TaskHound is the fix: one AI Desk for the whole team. Every team gets a shared TaskHound Container - a dedicated environment where all your projects, agents, and sessions live together. Fewer duplicated prompts, agent work you can actually review, and knowledge that persists from one session to the next.

Your whole team. One desk. Every agent.

TaskHound orchestrates your team's AI environment end to end - from the first prompt to the deployed review build.

One desk for the whole pack

Your team's Containers are secure, isolated environments where projects and agents live together. Every seat works in the same projects - same files, same branches, same knowledge - instead of five drifting laptop setups. TaskHound Containers →

Bring your own agents

Build a library of named agents backed by the Claude Code, Codex, Gemini, and Grok subscriptions and keys your team already pays for - each one personal to you or shared with the team. TaskHound never bills for AI usage. How pricing works →

See everything, everywhere

The live board follows you - web, iOS, Android, and native macOS and Windows apps. Push notifications when an agent finishes, fails, or needs input; approve from your phone. Multi-surface →

Push requests: humans hold the gate

Agents commit to THGit - TaskHound's central git host - and can never reach your real repo. When work is ready, they open a push request; a human reads the diff and approves before anything lands on GitHub, GitLab, or your own server. See the flow →

Teams · Containers · Projects

Structured like your org. Secured like production.

Your Team owns secure Containers - isolated environments your members get access to. Projects live inside them, and your agents run right beside the code: side by side, fully visible, never on anyone's laptop.

acme · Team 5 seats
production · Container Pack
▸ api
▸ checkout
staging · Container Scout
▸ data-jobs
▸ docs-site
  1. Your Team is the org.
    Seats, roles, and billing live here. Owners and admins see everything.
  2. Containers are secure environments.
    Isolated Ubuntu machines - provisioned by us, or bring your own. Snapshot-protected and access-controlled per member.
  3. Projects live in Containers.
    Real repos on real paths - shared by everyone with access, browsable and editable in the dashboard.
  4. Agents run beside the code.
    Claude Code, Codex, Gemini, Grok - securely inside the Container, visible to the whole team, never on a laptop.
No lock-in. No black box.

Everything lives in your TaskHound Container.

Projects, agent sessions, transcripts, and the team knowledge base all live together in one environment your team controls - not scattered across laptops. Open a terminal to it whenever you like, bring your own tools, and take your data with you anytime.

  • Full terminal and SSH access to your Container, always - it's yours.
  • Repos stay plain git checkouts; knowledge stays plain markdown. Leaving is one rsync.
  • Container snapshots before every agent session - roll a project, or the whole Container, back in one click.
acme - TaskHound Container 
$ ssh acme@container.taskhound.com
Welcome to your TaskHound Container - 4 agents on the trail

$ hound board
MARA   claude  api    feature/auth-refresh  running 25:44
DEVON  codex   web    fix/cart-totals       running 12:43

$ ls projects/api/.hound/knowledge/
auth-flow.md   deploy-runbook.md   query-cache-learnings.md

$ hound snapshots api | head -1
pre-session · 2 min ago · restore anytime
THGit · Push requests

Four layers of git between an agent and your repo.

Agent commits flow from the working tree through your Container's repo to THGit - TaskHound's central git host - and stop there. Nothing reaches your real repo until a human approves a push request.

  1. 1 · Worktree

    Agent commits its work

    Every session works in its own tree; sub-agents get isolated worktrees of their own and report back.

  2. 2 · Repo

    Lands in the TaskHound Repo

    Each Container keeps its own repo - the shared workbench where the whole team's agent branches live.

  3. 3 · THGit

    Syncs to THGit

    TaskHound's central git host. Clone or pull it like any remote - and browse repos, branches, and history in AgentGIT.

  4. 4 · Push request

    A human reviews

    The agent proposes a push; you read the diff and approve or decline. Merge conflicts? Hand them to an agent to resolve.

  5. 5 · Upstream

    Your repo, untouched by agents

    Approved pushes land on GitHub, GitLab, or your self-hosted git. Agents can never reach your upstream directly.

▸ The human gate is the point: agents propose, you read exactly what they committed, your team decides what ships. Want to see it running first? th-expose publishes any project's port at a live *.th.works preview URL.

Agent SSH · Tools

Give your agents reach - never your keys.

Agents on the Desk do more than edit files. They operate real systems and call real capabilities - while TaskHound holds every credential, so agents never see a secret.

Agent SSH: named Endpoints

Name the systems agents may operate on - servers, k8s boxes, the NAS - grouped in folders, each marked Development or Production. Production endpoints refuse destructive commands. TaskHound holds the SSH key and runs each command on the agent's behalf; the agent never touches credentials.

Relay: reach firewalled boxes

Box behind a corporate firewall or NAT? Install a small Go agent that polls outbound - no inbound port, no VPN, every request authenticated by Ed25519 signing. Same approval rules, same audit trail, anywhere the box can dial out.

A real toolbelt

Capabilities agents call as th-* commands: image generation and true image editing, background removal, 3D models, voice and sound effects, web search, and notifications. Your agent doesn't just describe the hero image - it makes it.

Web · iOS · Android · macOS · Windows

Watch your pack from anywhere.

Agents keep working after you close the laptop. TaskHound's board is live on every surface, and push notifications reach you the moment an agent finishes, fails, or needs your input - approve a pending command from the lunch line.

TaskHound - acme 4 live
MaraClaude Code

api · tests passed - 42/42 ok

TomGrokneeds input

infra · waiting for your approval…

Live demo: fix-cart-totals--acme.th.works published · 40s ago

TaskHound · now

Grok needs your approval

terraform apply · infra

Approve Open session

TaskHound · 4m

Mara's session finished ✓

3 learnings distilled · api

App Store Google Play macOS Windows Native apps - included with every seat.

How it works

Plan it, build it, review it, ship it - with a human gate before anything touches your repo.

  1. 01

    Plan together

    Sketch it in the Planning Yard, or start from a template - website, app, dashboard, game, service, CLI. Generate a plan, question it, convert it straight into a project.

  2. 02

    Agents build, you steer

    Live sessions, full transcripts. Steer mid-run, fork a session, or delegate to sub-agents in isolated worktrees. Task-board cards assigned to AI start their own sessions as capacity frees up.

  3. 03

    Review: AI, then human

    Work passes AI review before it reaches human review. Poke the running build at its live preview URL and read every commit in AgentGIT before you sign off.

  4. 04

    Approve the push, ship it

    A push request is the only road to your real repo - a human approves it, then it lands on GitHub, GitLab, or your own server. Agents never hold your upstream credentials.

The live board is the heartbeat.

The team board puts everything running on one screen: who's running what, on which branch, needing what - one glance. Steer your own sessions, pause anything as admin, and jump into any transcript mid-stream.

app.taskhound.com/board
⌘K  search sessions, knowledge, transcripts… 3 live · 1 needs input New session
Live demos checkout--acme.th.works docs-v2--acme.th.works
MK Mara Claude Code

api · feature/auth-refresh

editing auth/session.go

25:4448.2k tok

DV Devon Codex

web · fix/cart-totals

Steer Pause

12:4319.7k tok

TW Tom Grok needs input

infra · chore/tf-modules

▸ permission: terraform apply?
approve on any device →

07:118.9k tok

  1. 1Who's running what. Avatar, agent, project, and worktree on every card.
  2. 2Steer or pause. Owners steer mid-run; admins can pause or kill any session.
  3. 3Approve from anywhere. Pending permissions route to the board and push to your phone.
  4. 4Activity, not invoices. Token counts as a neutral work metric - your AI billing stays with your providers.

Simple pricing: seats + your Containers.

Pay per person, then size each Container by how many agents run at once. Your AI usage is between you and your providers - TaskHound never bills for AI usage.

Per seat

$19.95/user/mo

  • ✓ Team board, live sessions & steering
  • ✓ Per-project knowledge & full history
  • ✓ iOS, Android & desktop apps with push
  • ✓ THGit, push requests & th.works live demos
  • ✓ Stripe subscription - cancel anytime
Start your team

Your Containers - each one sized to how many agents run at once

Tier Specs Concurrent agents Live demos Snapshots Price
Scout 2 vCPU · 4 GB · 60 GB 2 1 7 days included
Pack popular 4 vCPU · 8 GB · 150 GB 5 3 30 days +$29/mo
Kennel 8 vCPU · 16 GB · 300 GB 12 10 90 days +$79/mo
Alpha 16 vCPU · 32 GB · 1 TB (GPU option) Unlimited 25 1 year +$199/mo

▸ Need just one more agent? Extra agent slot: +$4.95/mo on any tier - cheaper than a tier jump.

▸ Bring your own agents: your subscriptions, your keys. Linking providers is free and unlimited. Full pricing details →

Questions, answered straight.

Is my team's code isolated?

Yes. Each team gets its own dedicated TaskHound Container on fully isolated, dedicated infrastructure - nothing is shared with other teams, and no other tenant's workload touches your source. Container snapshots are taken before every agent session, so you can roll a project - or the whole Container - back in one click.

How do I pay for AI usage?

You don't pay us for it - ever. You bring your own agents: link the Claude Code, Codex, Gemini, or Grok subscriptions and API keys your team already has. TaskHound never resells or meters LLM usage and never shows you a cost ticker. You pay only seats ($19.95/user/mo) plus your Container tiers, as a Stripe subscription you can cancel anytime.

Which agents are supported?

Claude Code, Codex, Gemini, and Grok ship ready to link - and linking providers is free and unlimited. Every seat authenticates their own subscriptions in their own isolated space. All agents work side by side in the same projects and write to the same knowledge base, so you can pick the right agent per task.

What happens when all agent slots are busy?

New sessions queue, visibly, on the board - never a silent failure. You can wait, ask a teammate to wrap up, add an extra slot for +$4.95/mo, or upgrade the tier. Pricing attaches to concurrently running agents, so idle logins never count.

Can I follow my agents from my phone?

Yes - the board is live on the web, iOS, Android, and native macOS and Windows apps. Push notifications tell you the moment an agent finishes, fails, or needs your input, and you can approve or steer right from the notification.