Uncensored AI that doesn’t say no
The job: read Bob’s invoice from Alice’s account. Ablit ran the vulnerable and fixed builds in Kali. Before: the invoice leaked. After: access denied (403), with all five tests passing. See the finished work
SECURITY ASSESSMENT
A signed-in user could read another account’s invoices by changing the numeric id in the request. The endpoint checked that the caller was authenticated, but never that the invoice belonged to them.
| Finding | Severity | Status |
|---|---|---|
| IDOR on /api/invoices/:id | High | Fixed |
| Verbose error leaks stack trace | Medium | Fixed |
| Missing rate limit on login | Low | Open |
GET /api/invoices/4021 200 OK{ "account": "other-tenant", "total": "$18,400" }
Start a job from your phone. It runs in the cloud and keeps going after you lock the screen. Reopen the thread later to pick up the result.
Every run gets its own isolated cloud Kali box with the standard pentest tooling preinstalled. Ablit runs the commands and reads the output right in your chat — recon, scanning, exploitation and the retest.
Connect any MCP server in Settings and Ablit can call its tools during a run. Point it at your own server or a service you use, and tell it what you need in the chat.
Connect your toolsConnections use your own service access.
Connect Ablit to Cursor or any other IDE and use it in place of the general models.
Set up your IDE
https://ide.ablit.app/api/v1Real Cursor UI, with Ablit input annotations. Screenshot: Cursor Forum
ABLIT CODE
A coding agent on your own machine. It reads the project, edits files, runs the tests and asks before anything risky — with Ablit Flash, Ablit Max and the whole Ablit library behind it. One command to install, your API key to start.
$ curl -fsSL https://ablit.app/install | bash
Checksum verified. ABLIT CODE is installed as ablit.
$ ablit
┃ Connect Ablit · API key abl_live_•••••••••••• enter ↵
✓ Connected. Model: ablit/ablit-flash
> Add input validation to the signup form and cover it with tests
read src/signup.ts · src/signup.test.ts
edit src/signup.ts +38 −6
write src/signup.test.ts +54
run npm test → 12 passed
Done. Validation rejects empty names, bad emails and short passwords; every rule has a test.
Real jobs, start to finish
Security testing
Try to break my staging app. Can you get into another test account’s invoices? Show me how, patch the hole and run the test again.
The attack and the retest against a fixed build. See the before-and-after responses in the recording above.
Marketing & creative
Look at these three competitors. Find the gaps in their offers, write five ads for ours, and use ModelClone to make the images.
Browser research, ad copy, and image generation through your connected ModelClone account.
Research
Find a supplier for 500 printed boxes. Compare minimum orders, delivery times, and prices. Link the sources so I can check them.
Live web research with source links, organized into a comparison you can use.
Code & automation
These exports have different column names. Write a script to merge them, catch duplicates, and test it on the files I uploaded.
A script and its test output in the same thread. Save the instructions as a skill for the next batch.
Choose your plan
Plans include monthly credits. Buy more when you need them; top-ups stay until you use them.
Start with one connection
$20 / mo
For everyday tasks
$60 / mo
More credit and connections
$200 / mo
Monthly plan credit resets at renewal. Additional purchased credits stay until you use them. File storage supports up to 5 TB per upload on every plan; analysis depends on the file format, model context and workspace capacity.
Questions
Yes. Ablit runs on abliterated models, so it answers and does the work instead of refusing. It writes and runs code, pentests apps from its own Kali box, browses the web and uses your connected tools. The one hard line is sexual content involving minors.
Abliteration strips the refusal behaviour out of an open-weight model, so it completes a task instead of declining it. Ablit pairs that with a Kali workspace, a browser and tool connectors, so you get a finished job rather than a paragraph of caveats.
It does not refuse, and it runs the work end to end: code it writes and runs, its own Kali box, a browser, and your connected tools. Ask it to build something, break your test app, or do the research, and follow every step in the chat.
Sexual content involving anyone under 18, in any form. No setting or prompt changes that. Everything else is allowed, including adult content and offensive-security work on systems you are authorized to test.
Yes. Each account gets an isolated cloud Kali Linux box with persistent files and the standard security tools. The agent runs commands and reads their output. The system filesystem is read-only; your workspace, home and local-tool directories are writable.
Yes. It drives its own browser to navigate sites, sign in, fill forms, scrape data and click through workflows. You watch it happen from the conversation.
Add your HTTP or SSE MCP servers in Settings and the agent can call their tools during a run. Recon includes 1 connector, Operator 5, and Overlord 20. Install skills or playbooks, then tell the agent to follow one by name.
Yes. Once it starts, the job runs on the server, not in your browser. Lock your phone or close the tab, then reopen the same thread to see saved activity and results. Stopping a run cancels it; usage limits, run limits and provider errors can also end one.
Yes, on every plan. In an OpenAI-compatible Chat Completions client, set the base URL to https://ide.ablit.app/api/v1, paste your Ablit API key, and use ablit as the model. Usage draws from your Ablit credits. Your IDE supplies its project context and tools; cloud workspace access depends on your key’s permissions.
Offensive-security work is for systems you own or are authorized to test. Used that way, it is a standard pentest workflow. Anything beyond that is on you.