Chat thinks with you. Work gets it done.

Same agent, two gears. Chat is for decisions and drafts in the loop with you. Work takes the job onto its own computer, picks among 100+ models, and brings back the finished file.

GPT
Claude
Gemini
DeepSeek
Qwen
Kimi
GLM
Arya
Seedream
Wan
Kling
MiniMax
100+ models. The agent picks what the work needs.
New ones land on day one, at no extra cost, with nothing for you to migrate. Frontier, open-source, and Gab's own Arya. One workspace, one bill.

Its own computer. Not yours.

The agent has its own computer. Every job runs on an isolated machine in our cloud, created when the work starts, destroyed when the files are delivered. Its tools and the models it picks run there. Never on your laptop. Never on the same machine as another task.

A brand new machine every time a task starts
Its tools and the models it picks run there, not on your device
No access to your computer, your files, or your network
One task can never see another task’s work
Torn down the moment the finished files are delivered
Your computer
Your files stay where they are
Untouched
No access
Running in Gab AI’s cloud
TASK 1
Draft the Q3 update
fresh machine
TASK 2
Watch competitors
fresh machine
TASK 3
Build the launch kit
fresh machine
Three tasks, three separate machines. None of them can reach your device, and none of them can reach each other.

You ask or assign. It picks the right model.

The agent reads the intent and sends the step to whichever of 100+ models is genuinely best at that job. An image model for the poster. A video model for the teaser. A frontier LLM for the analysis. Pick by hand whenever you want.

Auto
Every request goes to the model that's best at it.
Image
GPT Image 2
Design a poster for our spring launch
Video
Veo 3.1
0:12
Animate it into a twelve-second teaser
Music
MiniMax Music
2:45
Score it with something warm and hopeful
Code
def forecast(rows):
df = frame(rows)
q = df["revenue"]
return q.rolling(3).mean()
Claude
Python
Model the revenue trend from this sheet

Its own tools. For every kind of work.

The agent does not work from a blank chat. It has image generation, research, decks, code, translation, and 100+ models behind those tools. Open one yourself when you already know the job, or let Work pick what it needs.

1
A dedicated tool for the job, not another blank chat
2
The right model already chosen for that kind of work
3
Use a tool yourself, or let the agent pick what it needs
Browse AI tools
The catalog
50+ live
1
Image Generator
Create
2
Image Editor
Edit
3
Blog Post Writer
Write
4
Research Assistant
Analyze
5
Code Generator
Code
6
PowerPoint Generator
Business
…and 44 more, each with the model and settings already chosen.

Set it once. Sit down to the file.

The same agent keeps going after you leave. Morning brief, weekly report, competitor watch, on its own computer, with the same tools and models, on its own cadence. The result is waiting in your chat.

Browse agents
Morning news brief
Daily · 7:00 AM
Running now
Delivered. Waiting in chat
Searched 14 sources
Read 6 top stories
Drafted the brief
Delivered to chat
34 runs delivered. Never missed

Speak the brief. Open the file.

Voice is Chat out loud. Walk through a decision while you drive, then tell the agent to draft the announcement and open the file. No button to hold. The transcript lands in the chat either way.

Talk over it mid-sentence and it stops to listen. Eight voices, on iPhone, iPad, Android, or the web.

How voice mode works
Speaking
02:14
“…so move the launch to the second week of October, and draft the announcement post while you’re at it.”
Cedar
Every session is transcribed into the chat it belongs to, searchable next to your typed conversations.

A real editor. Beside the chat.

The agent opens a report, deck, or budget in a real editor beside Chat. Tighten a paragraph yourself, or keep talking and watch the draft update.

1
Chat to rewrite a paragraph. Work to produce the whole file.
2
Upload existing Word, Excel, and PowerPoint files
3
Version history with one-click restore
Draft our Q3 investor update
Done. It’s open in the editor. Want a metrics table?
Yes, and tighten the intro
Updating the draft…
Ask for changes…
Export
Q3 Investor Update
Revenue grew 34% quarter over quarter, driven by self-serve upgrades and a full quarter of the new editor suite.
Net churn improved for the third straight quarter.
Metric
Q2
Q3
ARR
$4.2M
$5.6M
Net churn
2.1%
1.4%
Documents
Rich text with images, tables, and AI rewrites on any selection.
PDF · Word
Presentations
Structured slides you can reorder, restyle, and drag into place.
PowerPoint
Spreadsheets
A live grid with real formulas, ranges, and cell formatting.
Excel · CSV

Bring your history with you.

Switching shouldn't cost you years of context. Import the archive from ChatGPT, Claude, Grok, or another Gab AI account so the agent already knows the project. Every thread searchable, nothing retyped.

1
Drop in the export
We detect which service it came from and read the whole archive, ZIP or JSON, however many files it split into.
2
Threads arrive intact
Titles and timestamps come across, so your archive lands in order instead of as one undated pile.
3
Run it as often as you like
Conversations you have already imported are skipped, so a second pass only ever adds what is new.
How importing works
Import conversations
ChatGPT
conversations.json
48.2 MB
1,284
conversations imported
36
duplicates skipped
Oldest thread dated March 2023. Still searchable.
Exports we read
ChatGPT
conversations.json
Claude
Export Data ZIP
Grok
account data ZIP
Gab AI
account export

It remembers the project. Nobody else sees it.

Memory you control
Context you choose stays with the agent, so you stop re-explaining the same project. Edit or delete it any time.
Anonymous routing
Requests reach model providers without your identity attached. They never see who is asking.
Never training data
Your conversations, files, and creations are never used to train our AI model.

One API. Every model.

The same 100+ models the agent uses in Chat and Work, as one OpenAI-compatible endpoint. Point your SDK at it, swap GPT, Claude, Gemini, and the rest, or let Auto Router choose.

Swap any of 115 frontier and open-source models with one key
Pass model="auto" and Auto Router picks the best model for the task
Drop into Cursor, Claude Code, Codex, or any OpenAI-compatible SDK
Explore the API Router
from openai import OpenAI
client = OpenAI(
api_key="YOUR_GAB_API_KEY",
base_url="https://gab.ai/v1",
)
response = client.chat.completions.create(
model="auto",
messages=[
{"role": "user", "content": "Hello, how are you?"}
],
)
print(response.choices[0].message.content)

How your kind of work gets done

Writers, developers, churches, veterans, and more. Same agent, same tools, same 100+ models — aimed at the work your field actually does.

Loved by people who switched.

Real customers. Real words, from the App Store and from X.

For maximum truth-seeking, nothing beats Gab AI. Fast responses on the default model, with an incredible selection of models. This has been an AI gamechanger for me, allowing all chats to be stored in one convenient location.
Ethan Z.
App Store review
Gab AI is a breath of fresh air in the saturated AI space. The feature set of Gab AI is incredible. Repeatable tasks, custom memory, custom agents, voice interaction, and more. Unlike the rest of the industry, they do not use your conversations to train their models. Bottom line, this is the only AI that I would allow my kids to use.
Dr. Alan B.
App Store review
This is a SUPER useful app. I’ve used it for help with things like technical issues with cellphones and even complex health questions. I use another AI app as well, and when I ask it and Gab AI the same question, I get better answers from Gab AI.
Sarah T.
App Store review
On X

AI that works.

One agent with its own computer, its own tools, and 100+ AI models. Free to start. No credit card. Cancel anytime.

Start for free
No credit card
Cancel anytime