Hezo is a self-hosted AI agent team platform: one command gives you an entire AI workforce organized like a company. You define the goals and projects; your team of AI agents - CEO, Coach, Captain, engineers, designers, researchers - ships the work. Use the web app to manage projects, set budgets, chat with the CEO (or connect Slack, Telegram, and Discord), and track progress on a real task board.
The magic is in the teamwork. Each AI model runs where it works best - Claude, GPT, Gemini, or local models via Ollama and LM Studio - and Hezo provides the shared workspace where they collaborate: a task board, project documents, a shared skills database, and built-in quality checks that catch mistakes before work ships. You bring your own model provider accounts and control every dollar with daily, weekly, and monthly budget caps.
Your secrets stay yours. Agents reference API keys by placeholder - they never see the real values. The real key is only substituted at the moment of the request, and only for services you've authorized. Rotate a key from the dashboard without losing your provider setup โ edit it in place and Hezo verifies the new key before switching. Every project is isolated, and your data never leaves your infrastructure.
Over time, your AI team improves. Hezo's Coach reviews every completed task and feeds what it learns back into the team's skills and workflows. Teams can be exported as downloadable bundles and shared through the marketplace. Your team gets faster and smarter the more you use it.
Hezo is built for anyone who wants AI agents that get real work done - without handing over their secrets or their infrastructure. One binary. A Docker-compatible runtime is the only prerequisite - Docker, Colima, Rancher Desktop, OrbStack, and Lima all work - or use a managed sandbox service and skip the Docker setup entirely. Runs on modest hardware: Hezo automatically provisions swap space so it stays stable even on low-RAM VPSes.
A startup from Singapore that is founded by Ramesh Nair.
Agents organised like a company
a global CEO and Coach above a per-team Captain and workers, live CEO chat, automatic Coach reviews.
Structure the team to the work
compose the roster, the reporting lines, and what each role is, change it while a project runs, and carry a structure you've tuned forward to the next one.
Teams & projects
one team per project, launch a ready-made team from the marketplace or start from a template, hire and customize agents, snapshot a team to reuse, export one to share.
Autonomous task execution
a task board with sub-tasks nested up to three levels, heartbeat wake-ups, approvals, auto-resumed long runs.
Know where a project stands
a Progress page the Captain keeps current on its own, with optional high-level goals and scheduled re-checks layered on top.
One platform layer over every model
the meta-harness runs each model in its own first-party CLI, then levels the differences: the same tools, skills, memory and sandbox whichever you pick, plus an independent completeness check that won't let a run end on failing tests, an "out of scope" dodge, or a handoff nobody received.
Your models, your spend
bring your own providers, per-agent models, budget caps and cost tracking.
Choose where containers run
on your own machine or a managed sandbox service, switchable either way at any time, with containers started on demand and a memory budget shared across every project.
Secure by design
secret placeholders, encryption at rest, admin password sign-in, sandboxed containers, verified git commits, an audit trail.
Teams that improve themselves
the Coach writes durable learned rules back onto agents after each finished task.
Knowledge & memory
documents, skills, version history and restore, long-term chat memory, assets, full-text search.
Connect your tools, both ways
drive Hezo from any MCP client via its built-in MCP server, and give agents the services you already use with connectors โ hosted MCP servers or plain REST APIs โ scoped to one project or shared across all of them.
Chat from anywhere
run the CEO from Telegram, Slack, and Discord, as a private assistant or a coworker in your team channels.
Your data, in your storage
embedded Postgres, optional hosted Postgres, local or S3-compatible asset storage, data-preserving upgrades.
Speaks your language
the web app runs in 12 languages, picked from your browser on first run and set before anything else; date and currency formats are chosen independently.
Self-hosted & easy to run
a single binary that runs anywhere a Docker-compatible runtime does, or on a host with no container runtime at all when containers live on a managed service, one-click cloud-init, secure remote access, safe-rollback backups, in-app self-update, a mobile-first web app.
Hezo is the only self-hosted AI agent platform that runs each model inside its own native first-party harness - Claude Code for Anthropic, Codex for OpenAI, Gemini CLI for Google - rather than a lowest-common-denominator wrapper. That means each model uses the tooling it was actually designed for, so agents are more capable out of the box.
Security is structural, not bolted on: agents reference API keys by placeholder and never see the real values - substitution happens at the egress proxy only for services you've authorized. Every project runs in its own isolated container. All data is AES-256-GCM encrypted at rest behind a key only you hold.
Other differentiators: an independent completeness check that verifies work is actually done before a run ends (not just trusting the agent's self-assessment), a Coach that learns from every completed task and feeds improvements back into the team, per-agent budget caps with live cost tracking, and a real org-chart team structure (CEO โ Coach, Captain โ engineers) rather than a flat list of agents. One command to install. Your infrastructure, your model keys, your data.
Developers and AI/ML engineers who want AI agents that get real work done - without handing over their API keys, their infrastructure, or their data. The primary audience is technically capable, already comfortable with self-hosting and the command line, and values data sovereignty and cost control. They range from solo developers who need leverage without hiring, to small teams running their own AI workforce on modest hardware.
Hezo was created by to solve a problem: running teams of AI agents that do real work requires keeping secrets actually secret, controlling costs, and independently verifying that work is complete โ not just trusting the agent's word. The name is a play on hรฉzuรฒ (ๅไฝ), Mandarin for "collaborate/cooperate." The platform launched in March 2026 and has been shipping steadily since.
TypeScript throughout, running on Bun. The web UI is React + Vite. The server, API, realtime engine, embedded database, secret vault, egress proxy, and Docker orchestration are all compiled into a single self-contained binary. The monorepo uses Turbo for orchestration and Biome for linting/formatting. Testing: Vitest + Playwright. Container isolation: Docker (compatible with Colima, Rancher Desktop, OrbStack, and Lima).
Hezo is early-stage and self-hosted โ users run it on their own infrastructure, so we don't have a public customer list or named enterprise logos. The project is source-available on GitHub (github.com/hezo-ai/hezo), growing actively, and used by developers who value keeping their data and secrets on their own machines.
We have collected here some useful links to help you find out if Hezo AI is good.
Check the traffic stats of Hezo AI on SimilarWeb. The key metrics to look for are: monthly visits, average visit duration, pages per visit, and traffic by country. Moreoever, check the traffic sources. For example "Direct" traffic is a good sign.
Check the "Domain Rating" of Hezo AI on Ahrefs. The domain rating is a measure of the strength of a website's backlink profile on a scale from 0 to 100. It shows the strength of Hezo AI's backlink profile compared to the other websites. In most cases a domain rating of 60+ is considered good and 70+ is considered very good.
Check the "Domain Authority" of Hezo AI on MOZ. A website's domain authority (DA) is a search engine ranking score that predicts how well a website will rank on search engine result pages (SERPs). It is based on a 100-point logarithmic scale, with higher scores corresponding to a greater likelihood of ranking. This is another useful metric to check if a website is good.
The latest comments about Hezo AI on Reddit. This can help you find out how popualr the product is and what people think about it.
Do you know an article comparing Hezo AI to other products?
Suggest a link to a post with product alternatives.
Is Hezo AI good? This is an informative page that will help you find out. Moreover, you can review and discuss Hezo AI here. The primary details have been verified within the last quarter. So they could be considered up to date. If you think we are missing something, please use the means on this page to comment or suggest changes. All reviews and comments are highly encouranged and appreciated as they help everyone in the community to make an informed choice. Please always be kind and objective when evaluating a product and sharing your opinion.