Coldtea.ai is a desktop IDE that bundles a multi-agent terminal, automated QA agents, and production monitoring into one workspace, so a team running Claude Code, Codex, or OpenCode doesn't have to stitch those pieces together by hand. It launched publicly on Product Hunt on August 7, 2026, and reached the top spot on the site that day.
Short answer: Coldtea.ai is a desktop app that runs coding agents (Claude Code, Codex, Gemini, OpenCode) in a shared terminal, then adds visual QA agents and production monitoring on top. It's free to start — 2,000 agentic credits a month, roughly 200 test runs — with paid seats at $20/user/month and an optional self-driving add-on from $49/month per project.
When I tested the setup flow by working through Coldtea's own docs, its live pricing page, and its Product Hunt listing on launch day, there was no track record yet to lean on beyond what the company publishes itself. This is what the setup flow actually asks of you, what each pricing tier includes, and where I'd slow down before connecting a real repository.
What you'll need
Coldtea is a native desktop app, currently distributed as a macOS build (I pulled the download link directly and it resolved to an Apple Silicon .dmg, version 1.0.13 at the time of writing — check the download page yourself if you're on an older Intel Mac or a different OS). You'll also need at least one coding agent CLI already installed and authenticated on your machine, since Coldtea doesn't install agent binaries itself; it only detects what's already on your shell path. A local Git repository is the other requirement — Coldtea's setup flow ends with picking or creating one as your working directory. If you want cloud-based runs or the QA and monitoring agents, you'll also connect a GitHub account and either an API key or a subscription for whichever agent you use.
Step-by-step: setting up Coldtea.ai
1. Download and install the app
Grab the desktop build from coldtea.ai and move it into your Applications folder if macOS asks. There's no separate installer step beyond that — launching the app starts the first-run flow.
2. Sign in and create a TeaHouse
Coldtea signs you in with a Google desktop sign-in, then asks you to create what it calls a TeaHouse — its term for a team workspace — by naming it and optionally adding a website URL. This is step 1 of 4 in the on-screen setup counter.
3. Connect GitHub
Step 2 authorizes GitHub access so cloud and background agents can open pull requests on your behalf. Skip this if you only plan to run agents locally, though you'll lose the cloud task and PR-creation features.
4. Set your agent credentials
Step 3 asks you to choose between an API key or a subscription connection for Claude, Codex, or OpenCode, depending on which agent you plan to run in the cloud. Local-only use skips this — the CLI you already have installed authenticates itself the normal way in its own pane.
5. Pick a project
Step 4 opens an existing Git repository folder or lets you create a new one. In my check of the docs, Coldtea explicitly designs this flow to be resumable — if you've already completed a step in a prior session, it detects that and skips ahead.
6. Open the terminal and run an agent
Once a project is loaded, open a new pane and pick a detected agent from the picker. If nothing shows up, that's a local install or PATH problem, not a Coldtea bug — run claude --version (or the equivalent for your agent) in your normal shell first to confirm it's reachable.
Example prompts you can copy
Coldtea doesn't run its own model — you're prompting whichever agent CLI you connected — but a few starting points work well once you're inside the terminal pane:
- "Review the open PRs assigned to me, summarize what changed, and flag anything that touches the auth module."
- "Run the test suite, then open a background cloud task to fix any failures without waiting for me."
- "Draft a QA scenario in plain English: log in, add an item to the cart, and confirm the total updates correctly" (this is the kind of instruction the Visual QA agents are built to convert into a self-healing test).
- "Summarize what changed in production since the last deploy and flag anything that looks like a regression."
Specific, scoped requests work better than open-ended ones — a vague "fix the bugs" prompt gives a background agent nothing to check its own work against.
Common mistakes to avoid
The first mistake I'd flag: assuming Coldtea installs your coding agent for you. It doesn't — if Claude Code or Codex isn't already on your machine and on the shell path Coldtea launches from, the agent picker just won't show it, and that's easy to mistake for a bug. Second, connecting a production repository before testing the flow on something disposable; the self-driving and monitoring features are the parts most likely to open pull requests or take actions on your behalf, so a throwaway project is the safer place to learn the interface first. Third, ignoring the credit math on the free tier — 2,000 agentic credits works out to roughly 200 test runs a month, per Coldtea's own pricing page, which is easy to burn through faster than expected once QA agents are running against every PR. Fourth, treating the $20/user/month Pro price as the full cost: the self-driving add-on ($49–$99/month per project) and the Agentic Testing Pro add-on ($120/month for 10,000 credits) are separate line items, not included by default. Fifth, skipping GitHub connection and then being confused when background cloud agents can't open PRs — that step is what makes the cloud and monitoring features work at all.
Coldtea.ai pricing at a glance
| Tier | Price | Includes |
|---|---|---|
| Free | $0/forever | Unlimited synced tasks (Linear, Jira), any CLI agent, 2,000 agentic credits/month (~200 test runs), 2 connected tools/project, 7-day self-driving trial per project |
| Pro | First seat free, then $20/user/month | Everything in Free, shared boards, optional Agentic Testing Pro add-on, unlimited connected tools on self-driving projects |
| Self-driving add-on | $49–$99/month per project | Essentials or Complete tier, added on top of Free or Pro |
| Agentic Testing Pro | $120/month | 10,000 credits, roughly 1,000 test runs |
| Enterprise | Custom | Volume credit pricing, annual billing, priority support, shared Slack channel |
I confirmed every figure on this table against Coldtea’s own pricing page on August 7, 2026 — the day it launched. Usage-based pricing on a same-day launch tends to move, so check the live page before budgeting a team around it.
Tools that make this easier
If you're evaluating Coldtea against what you're already running, my best AI tool for code roundup compares it to standalone editors and copilots rather than a multi-agent orchestration layer. For the terminal-first coding agents Coldtea is built to sit on top of, my honest review of AI programming tools covers how well agent-written code holds up without a human in the loop. If Claude Code specifically is your primary agent, how usage tracking works with Claude Code is worth reading alongside Coldtea's own credit system, since you're now managing two different usage meters. Anyone running several agents in parallel and worried about them stepping on each other's changes should also see this local merge queue for parallel Claude Code agents, which solves a related coordination problem from a different angle. And if ChatGPT-based tooling is more your stack than Claude or Codex, ChatGPT alternatives for programming and my broader AI coding assistant guide cover where those fit instead.
My take
Coldtea's pitch — one workspace for the terminal, QA, and monitoring instead of three separate tools bolted together — solves a real coordination problem for teams already running multiple coding agents. The free tier is generous enough to actually test the workflow before paying, and the pricing page is unusually specific for a same-day launch, which made it easier to verify than most. What I can't tell yet, because the product is brand new, is how the self-healing QA agents and the production monitoring hold up against a real app under real load — that's the part worth trialing on a low-stakes project before you let it open pull requests on anything that matters.
Last updated: August 7, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
Frequently Asked Questions
Is Coldtea.ai free?
Yes, to start. The Free tier is permanent and includes 2,000 agentic credits a month (about 200 test runs), any CLI agent, and a 7-day self-driving trial per project, with no credit card required. Paid seats begin at $20/user/month once a team wants shared boards and unlimited connected tools.
How long does it take to set up Coldtea.ai?
The four-step first-run flow — TeaHouse, GitHub, agent credentials, project selection — takes a few minutes if you already have a coding agent installed and authenticated locally. Getting real value out of it takes longer, since you still need to configure the QA and monitoring agents against your actual app before they're catching anything useful.
What is the easiest way to start with Coldtea.ai?
Install the app, connect one agent you already use locally, and run it against a throwaway or low-stakes repository first. Skip GitHub and the self-driving add-on until you've confirmed the basic terminal and agent picker work the way you expect.
Does Coldtea.ai replace my coding agent?
No. Coldtea doesn't run its own model — it's an environment for the agent CLIs you already use (Claude Code, Codex, Gemini, or OpenCode), plus QA and monitoring layered on top. You still need one of those agents installed and authenticated separately.
What platforms does Coldtea.ai support?
At launch, the distributed build is a macOS app (the download link I checked pointed to an Apple Silicon .dmg). Coldtea's docs don't list Windows or Linux desktop support as of this writing, so confirm current platform availability on coldtea.ai before assuming it runs on your setup.