Sandboxed by default
Each agent runs in its own isolated Docker container and touches only the directories you allow.
Apprentice · Local · Private · No telemetry
Run AI agents on your own machine.
Apprentice is a desktop app for building your digital team: local, private agents that browse the web, work with files, run tools, and hand work to each other. No cloud middleman, no per-run fees. Bring the AI subscriptions, cloud providers, or local models you already use.
- 30-day full Pro trial
- No account needed
- Nothing phones home
Downloads
Early days, real downloads.
Apprentice is young and already putting agents to work. Here is the live count, straight from GitHub.
- 825 Total downloads
- 325 Windows
- 273 macOS
- 227 Linux
GitHub download stats · no telemetry · updated hourly
Not a chat box. Apprentice agents actually use your computer, driving a browser, reading and writing files, running shell tools, and calling your MCP integrations. Real work, on a real machine.
Set it up
Hire an agent in minutes.
Give each one a purpose, a model, memory, a budget, a schedule, and exactly the tools it needs. No wiring, no glue code.
Teamwork
One agent hands work to the next.
Agents coordinate through chats, mentions, tasks, and shared activity, and every run is visible, so you always know what happened and why.
Private by design
Nothing phones home.
Apprentice runs on your machine: your data, your credentials, your model providers. The Community edition sends nothing at all. Pro makes exactly one network call, checking your license is valid. That's it.
Even downloads are just anonymous counters, with no account, no cookies, and no personal data. How I handle data.
Use cases
A teammate for the work that lives in tabs, inboxes, and dashboards.
-
Support
Triage requests, draft replies, and create follow-ups.
-
Research
Watch sites and feeds on a schedule and report what changed.
-
Community
Answer repeat questions and summarize feedback.
-
Sales
Research accounts, draft outreach, and track follow-ups.
-
Engineering
Run focused triage, release, or DevOps agents with scoped access.
-
Personal
Handle reminders, weekly reviews, and recurring admin.
A few examples, not a menu. If you can describe the job, you can build an agent for it. Your needs are the only limit.
You stay in control
Autonomy, inside boundaries you set.
Budgets that hold
Per-agent spending caps with daily, weekly, or monthly resets. You choose what happens on overrun: pause, notify, or hard stop.
Guardrails and permissions
AI-moderated guardrails, command allow and block lists, website filtering, and a permission broker for sensitive actions.
Everything on the record
Every run is logged on your machine. See what each agent did, when, and why.
Bring your own AI
Works with the models and channels you already use.
Providers
- Claude Code
- Codex
- Antigravity CLI
- ChatGPT Subscription
- Anthropic / Claude API
- Google / Gemini API
- OpenAI
- xAI (Grok)
- DeepSeek
- Mistral
- Kimi (Moonshot)
- GLM (Zhipu)
- Qwen (Alibaba)
- Groq
- Cerebras
- Fireworks
- Together.ai
- OpenRouter
- NeuralWatt
- OpenCode Go
- Kilo Gateway
- Ollama
- Ollama Cloud
- LM Studio
- Docker Model Runner
API keys, the AI subscriptions you already pay for, or fully local models. No markup on model calls. Image generation included, with Nano Banana via Google Gemini and GPT Image and DALL-E via OpenAI.
Triggers & channels
- Telegram
- Discord
- Slack
- Signal
- Twitch
- Cron schedules
Message an agent from your phone or put it on a schedule. It works while you don't.
Why local
Cloud platforms charge per task, forever. Apprentice doesn't.
| Apprentice | Cloud agent platforms | |
|---|---|---|
| Where agents run | On your machine, sandboxed in Docker | On someone else's servers |
| Pricing | Free, or a flat subscription with no per-task fees | Per task, per run, or per seat, forever |
| Your data | Never leaves your computer | Passes through their cloud |
| Model costs | Your keys and subscriptions, no markup | Bundled credits with a margin |
| What agents can do | Use a real computer: browser, files, shell, MCP | Whatever connectors they offer |
Pricing
Start free. Upgrade when your agent team grows.
Every install starts a 30-day full Pro trial. Every paid tier includes a 30-day money-back guarantee.
Community
Free forever
The full product for one agent at a time. No account, no telemetry, no time limit.
- Up to 3 agents, 1 active at a time
- 1 run at a time
- Every integration, guardrail, and budget control included
- Local models, cloud APIs, or your AI subscriptions
Pro
€99 €59 / year
Founding price for early customers, locked in for as long as you stay subscribed. That's €4.92 a month, billed yearly.
€9.9 / month
Billed monthly, cancel anytime. Switch to yearly to lock in the €59 founding price.
Your whole agent team, working in parallel.
- Unlimited agents, all running in parallel
- Use on 2 devices, up to 2 device changes a year
- Perpetual fallback license: cancel and keep the versions you paid for
- Direct support from me
Lifetime Limited
€249 one-time
A small, fixed batch of seats. When they're gone, this closes for good.
Pay once, own Pro forever, and fund focused development.
- Everything in Pro, plus every future feature
- Use on 2 devices, up to 2 device changes a year
- 1 year of direct support
- Lifetime member badge and roadmap priority
Pricing questions
What is the founding price?
The list price for Pro is €99/year. While the founding offer lasts, early customers pay €59/year, and that price is kept for as long as the subscription stays active. No silent hikes. If you cancel and come back later, you pay the price at that time.
What happens if I cancel Pro?
Yearly Pro includes a perpetual fallback license: after your paid year, you keep using the newest version released while you were subscribed, forever. You only lose access to newer updates and direct support. Monthly Pro does not include the fallback license.
How does the free trial work?
Every install starts with 30 days of the full Pro experience, no card required. When the trial ends, Apprentice drops to the Community edition. Nothing breaks and your agents and data stay on your machine.
What is the refund policy?
Every paid tier has a 30-day money-back guarantee. I will ask one question, what did not work for you, so I can make Apprentice better. Your refund does not depend on the answer. It covers a lifetime purchase and the first payment of a new subscription. Renewals are not refundable, but you can cancel anytime before the next billing date.
Is the lifetime tier really limited?
Yes. It is a single small batch of seats, sold once. When they are gone it closes permanently, with no second wave and no raised cap. Lifetime covers every future Pro feature.
How many devices, and what about VAT?
Each license runs on 2 devices at once, and you can change devices up to twice a year, so a license stays with you rather than being shared across a rotating set of machines. Payments are handled by Paddle as the merchant of record; VAT or sales tax is calculated at checkout based on your country.
Download
Latest version
Apprentice v2026.07.04
From the maker
I'm Ilgit, and Apprentice is a one-person product on purpose. No investors pushing per-seat pricing, no growth team, no telemetry dashboards. I run my own agents with it every day, and I answer support tickets myself. If it doesn't earn its place on your machine, there's a 30-day refund.
Blog
Latest posts
Dario Amodei Said Open Source AI Was a "Dangerous Path." Three Years Later, He Is Still Wrong.
Amodei's 2023 Senate testimony calling open source AI dangerous resurfaced in 2026. Reddit's technical community fact-checked every claim and found the same pattern that began with GPT-2 in 2019: safety language that happens to benefit Anthropic's bottom line.
An AI Passed Peer Review. The Team's Own Fix Proves the Problem Is Real.
Sakana AI's AI Scientist scored higher than 55% of human submissions at ICLR 2025. Each paper cost $15. Nature published a paper about the system. The team's recommended fix is an admission the existing system cant tell the difference.
Fugu Ultra Is an Orchestrator. The Benchmarks Are Someone Else's.
Sakana AI's Fugu Ultra benchmarks match Anthropic's best model. But the scores belong to GPT-5 and Claude Sonnet 4. Here is what Fugu actually is, and what it isnt.