An IDE where the agent writes and checks the code itself
Describe the task in words. The agent will read the project, make edits, run tests and open the page in the built-in browser. Tokens for Claude, GPT and Gemini below official prices.
The IDE is free, you pay only for tokens50,000 tokens on sign-upUp to 10× cheaper than official prices
Slafy IDE
How it works
Set a task like you would for a colleague, get a finished result
No need to copy code into a chat and back. The agent finds the files it needs, and you see every step it takes.
01
Describe the task
“Add a signup form and check it in the browser.” The file open in the editor can be attached in one click.
02
The agent works in your project
Reads files, writes edits, runs commands. Every step is visible in the chat, and edits are shown as a diff.
03
Checks and reports back
Runs tests, opens the page in the browser and briefly writes what was done and how it was verified.
Under your control
No edit reaches a file without your “Apply”
The agent proposes a change, and the chat shows which file it's changing and how many lines it added and removed, just like on GitHub. Click the file to open the diff: removed lines in red, new ones in green. You decide whether to apply or reject it.
“Apply” right in the chat, click a file to open its diff
Commands wait for confirmation too, with the full text visible in advance
The agent takes a rejection into account and offers another option
Slafy IDE
Built-in browser
The agent opens the site itself and checks that everything works
The browser is built right into the IDE. The agent sees the page text, clicks buttons, fills in fields and reads the console, while you watch the same page next to the chat.
localhost and local addresses open without asking
External sites only with your permission, and every click too
The agent sees console errors and fixes them itself
Slafy IDE
“Goal” mode
The agent keeps working until the task is done and verified
Choose “Goal” in the mode menu or start your message with /goal. The agent doesn't stop halfway: it checks the result against the task, runs tests and looks at the page in the browser.
Up to 200 steps without asking you anything
Up to 8 checks against the task items before answering
At the end, a list of what was done and how it was verified
Slafy IDE
Check for inconsistencies
A second pair of eyes in one click
The “Check for inconsistencies” button asks the model to reread its answers and edits and compare them with the project files. It looks for contradictions, errors and unverified claims, and doesn't change anything itself.
The check only reads: no edits or commands
Findings as a list, with how to fix each
The same button is in the website chat
Slafy IDE
More screenshots
Small things that make the IDE pleasant to work in
Slafy IDE
Ultracode and queue
While the agent is working, keep typing: messages are queued and sent one at a time. You can remove the ones you don't need, and after a stop the queue waits until you resume it.
Slafy IDE
Out of tokens
Instead of an error, the IDE shows a clear banner with a “Top up” button. Pay on the website, and the tokens show up in the IDE without a restart.
Slafy IDE
Model and settings
A Slafy account or your own API, model choice with the price per 1M tokens, context compression and RAG memory for a specific project.
Slafy IDE
Sign in in a minute
Click “Sign in via website” or sign in with your slafy.co username and password. New users get 50,000 tokens right away.
Features
What else Slafy IDE can do
Any model
Claude, GPT, Gemini, Grok, DeepSeek, Kimi and GLM. Pick the model above the chat: give hard tasks to a strong one and routine to a cheap one.
Reasoning effort
“Fast”, “Normal”, “Deep” and “Ultracode”. At the maximum, the model reasons longer, and the agent takes more steps and checks its edits with tests.
Auto mode
Edits are applied without the “Apply” button, tests and builds run on their own. Dangerous commands, .env and package.json still ask first.
The connection drops, the task doesn't
Up to 10 reconnects in a row. The agent continues from the same step, no need to explain anything again.
Spending in plain sight
Under each answer you see how many tokens the task used. After a top-up, the balance updates in the IDE on its own.
Chats by project
Each folder has its own history. Chats and drafts survive restarts and IDE updates.
RAG memory
For large projects: the IDE remembers rules and decisions in the memory folder and feeds them to the model before a task. Turned on with a checkbox.
Your own API
An OpenAI key or any compatible server instead of a Slafy account. Handy for local models and corporate gateways.
No prying eyes
No telemetry, and the editor doesn't load from the internet. Only the model you chose gets your code.
Price
The IDE is free. You pay only for tokens
No subscription. You top up your balance, and it's spent only on model requests. The same balance works in the website chat and the API.
macOS. The build isn't signed by Apple yet. Drag Slafy IDE to Applications and open it. If macOS blocks it, go to System Settings → Privacy & Security and click “Open Anyway”.
Windows. If SmartScreen warns about an unknown publisher, click “More info” → “Run anyway”.
IDE questions
How much does Slafy IDE cost?
The IDE itself is free, with no subscription. You only pay for model tokens at the pricing rates, and you get 50,000 tokens on your balance when you sign up. Under every agent answer you see how many tokens the task used.
Which systems does it run on?
macOS on Apple Silicon (M1 and later) and Intel, Windows 10 and 11 (64-bit). No Linux version yet.
Can I use it without a Slafy account?
Yes. On the sign-in screen, click “Connect via API” and enter your OpenAI key or the address of any compatible server. You won't need a Slafy balance then.
Can the agent break something?
The agent reads and edits files only inside the open project folder. By default, every edit is saved only after you click “Apply” (click a file to open its diff), and commands run after you confirm them. In auto mode, file edits and safe commands go through without asking: tests, builds, linters, git status and diff. Other commands, as well as .env, package.json and repository settings, still wait for confirmation.
What can the agent do in the browser?
Open pages, read text, click buttons, fill in fields and check the console. Local addresses like localhost:5173 open right away. The agent opens an external site only with your permission and asks before every click and text input. Downloads and access to the camera, microphone and location are disabled in this browser.
What happens if the internet drops?
The IDE reconnects up to 10 times and continues the task from the same point: the context and completed steps are kept. If the connection doesn't come back, you can continue the task manually.
Where does my code go?
Only to the model you choose, via slafy.co: your messages and the files the agent read for the task. The IDE collects no telemetry and doesn't load the editor from the internet.
How do I update the IDE?
When a new version comes out, the IDE shows a message with a download link. Chats and settings are kept after the update.
How do I contact support?
Message us on Telegram: t.me/Sawasdee_krap. There's also a support link inside the IDE, in the bottom bar.
