NXT AIBeta
Every model your work needs, in one workspace.
NXT AI is an AI operating environment: sessions that remember, tools that finish the job, a model selector that knows your plan, a shared prompt library and usage you can see — instead of another chat window.
- 6 models · 3 tiers
- 5 tools + model compare
- One API for all of it
New session
2 model tiers on the Pro plan · answers render as Markdown
Conversations
- Q4 onboarding email sequence2h
- Idempotency keys vs request IDs9h
- Warehouse pilot interview notes1d
Results
Text
INVOICE INV-2041
Northwind Supplies · Issued 28 Sep 2026
Total due $4,812.50 · Due 28 Oct 2026
Fields to extract
invoice number, supplier name, date, total
Result
Claude Haiku 4.5 · 270 tokens
{
"invoice_number": "INV-2041",
"supplier_name": "Northwind Supplies",
"date": "2026-09-28",
"total": "4812.50"
}Model A
Llama 3.3 70B
184 tokens
The second headline is clearer: it names an outcome customers want instead of listing features, and it scans in one glance.
Model B
Claude Haiku 4.5
212 tokens
Headline B. “Close the books in a day” is concrete and measurable; A relies on abstract adjectives a B2B buyer will skip.
This month
24% of this month’s tokens used
Models on the Pro plan
- Llama 3.1 8B (fast)
- Gemma 4 26B
- Claude Haiku 4.5
- Llama 3.3 70B
- Claude Sonnet 5.5
- Claude Opus 5.5
Routing
Each request finds the right model
Cost-aware routing is built in, so a quick translation doesn’t use the same model as a 40-page contract review.
- 01
You ask
A question, a document, or a tool like Summarize or Extract.
- 02
Pick a tier
Fast, Smarter or Most capable — your plan decides which tiers you can use.
- 03
Cache & shortcuts
Repeat work comes from your workspace’s cache; trivial inputs skip the model entirely.
- 04
Model answers
Routed to an available model in that tier, across multiple providers.
- 05
Usage is counted
Requests and tokens land on your usage meters, against the plan’s allowance.
Tools
Tools that finish the job
Each tool is a tuned prompt with validation, so you get the result — not a conversation about it. Very short inputs are answered without a model and cost nothing.
Summarize
Sample input: 1,240-word board update
Sample output: • Revenue up 18% quarter on quarter • Two senior hires closed • Pilot extended to a second site
Translate
Sample input: Your order has shipped and will arrive on Thursday.
Sample output: Su pedido ha sido enviado y llegará el jueves.
Rewrite
Sample input: We regret to inform you the feature will be late.
Sample output: A quick update: this feature needs a little more time.
Proofread
Sample input: Their going to recieve it on monday.
Sample output: They’re going to receive it on Monday. Changes: 3
Extract data
Sample input: Invoice INV-2041 · total $4,812.50
Sample output: { "invoice_number": "INV-2041", "total": "4812.50" }
Sample inputs and outputs.
Compare & choose
Two models, one prompt, side by side
Send the same prompt to two models and read both answers with their token counts. Find the cheapest model that’s good enough — then use it.
- Each model counts as one request
- Only models your plan includes are offered
- Higher tiers stay visible, marked “Upgrade”
Model A
Llama 3.3 70B
184 tokens
The second headline is clearer: it names an outcome customers want instead of listing features, and it scans in one glance.
Model B
Claude Haiku 4.5
212 tokens
Headline B. “Close the books in a day” is concrete and measurable; A relies on abstract adjectives a B2B buyer will skip.
For teams
A prompt library and usage you can see
Save the prompts your team writes again and again and share them with the workspace. Requests and tokens are metered against your plan, with daily and monthly limits that reset on schedule.
- Personal or workspace-shared prompts
- Requests and tokens per day, per model
- Chats are private to the person who started them
This month
24% of this month’s tokens used
Models on the Pro plan
- Llama 3.1 8B (fast)
- Gemma 4 26B
- Claude Haiku 4.5
- Llama 3.3 70B
- Claude Sonnet 5.5
- Claude Opus 5.5
API
The same tools, over HTTPS
Create an API key with the ai:invoke scope and call every task from your own code. Available on plans that include API access.
curl -X POST https://nxtcloud.app/nxt-ai/api/v1/run \
-H "Authorization: Bearer $NXT_API_KEY" \
-H "content-type: application/json" \
-d '{"task":"translate","input":"Good morning","option":"Spanish","tier":"economy"}'{
"output": "Buenos días",
"model": "llama-8b",
"tier": "economy",
"cached": false,
"usage": { "input_tokens": 31, "output_tokens": 6 }
}Tasks: chat, summarize, translate, rewrite, proofread, extract. Example response values are illustrative.
Honest by design
What NXT AI does with your work
Private to you
Conversations belong to the person who started them; prompts are shared only when you choose.
Cache stays in your workspace
Cached answers are keyed to your organization and never served to anyone else.
Export or delete
Your chats, prompts and tool results are included in your account export and removed when you delete your account.
AI can be wrong
NXT AI isn’t a lawyer, doctor or financial adviser. Check important information.
Plans
NXT AI plans
Priced per workspace. Annual billing costs ten months instead of twelve.
See every limit and compare plans on the pricing page.
Ecosystem
Works well with
Same account, same workspace, no extra sign-up.