One workspace.
Every model. No decisions.
Nitrah reads what you are actually trying to do and sends it to the AI best suited to the job — then shows its work. You never pick a model again, unless you want to.
Free plan, no card required. Credits reset monthly.
- Understanding request
- Matching capabilities
- Selecting model
Routing across
- Providers
- 3
- Models
- 9
- Capabilities
- 10
Counted live from the model registry, not a marketing figure.
The problem
Choosing the model is not your job
Every provider is best at something and worse at something else, and the answer changes every few months. Keeping up means reading release notes instead of doing your work — and paying premium prices for questions that never needed a premium model.
Nitrah turns that into a routing problem and solves it once, for everyone. A registry describes what each model can do, what it costs and how fast it is. A router matches your request against it. When a better model appears, it becomes a row in the registry, not a habit you have to relearn.
- Capability matched
- Vision, tools, reasoning and context are checked before a model is chosen — never assumed.
- Priced before it runs
- Credits are reserved up front and settled against real provider usage, never a client-side number.
- Fails over safely
- Only to a model that can actually do the job, and only before you have seen any text.
- Always inspectable
- Every answer shows which model produced it and whether a fallback was used.
Capabilities
A workspace, not a collection of tabs
Everything below shares one context, one billing account and one set of permissions — and every one of them is built and running today. Nothing here is a mockup.
Chat that picks its own model
Ask anything. Nitrah reads the request, routes it to the model best suited to it, and streams the answer back — telling you which model answered.
Automatic failover
When a provider rate-limits or goes down, a capability-compatible model takes over and you are told it happened. No silent downgrades.
Cost-aware routing
Every request is priced before it runs and metered after. Simple questions go to fast, cheap models; hard ones earn the expensive model.
Documents and data
Upload PDFs, spreadsheets and documents. Ask for risks, summaries, comparisons — answered from the passages that actually matter, with locations cited.
Web research
Research mode searches, reads and compares sources, then answers with real citations. Anything it could not verify is marked as unverified.
Image studio
Generate, edit and iterate on images in a workspace that keeps your history, prompts and variations together.
Audio and voice
Transcribe recordings, talk to your workspace, and turn answers into speech where the provider supports it.
Agents that finish the job
Give a goal, not a prompt. Agents plan, use tools and verify their work inside hard limits on steps, time and spend that you control.
Projects with memory
Group chats, files and instructions into a project. Context carries across every conversation inside it, so you stop re-explaining yourself.
How it works
New models are configuration, not a rewrite
Providers sit behind one interface. Models, prices, context limits and fallback chains live in the database, so adding a model released this morning is an admin task — not a deploy.
- 01
Understand
The request is analysed for capability, size and reasoning depth.
- 02
Select
Candidates are filtered by capability, context and your plan, then ranked.
- 03
Reserve
Credits are held against your account before a single token is spent.
- 04
Stream
The answer streams back, and usage settles against what the provider reported.
Pricing
One subscription. Every model.
Usage is metered in credits, so a short question costs far less than a long analysis. You always see what a request will cost before it runs.
Pro
20,000 credits per month
- Document analysis
- Web research with citations
- Image studio
Credits are consumed in proportion to what each request costs to run, measured server-side from what the provider reports. Cancel any time — no contract, no exit fee.
Security & privacy
Boring, on purpose
Your workspace is the authorization boundary. Every request that touches a conversation, file or credit balance re-checks your membership of that workspace server-side — there is no code path that reads a record by id alone.
Keys stay server-side
Provider API keys are read from the server environment and never reach the browser.
Checked on every request
Authorization is enforced per resource, not per page. Plan status is read from our database, never from the client.
Usage computed server-side
Token counts come from the provider response. Nothing billable is ever accepted from the client.
Your data, deletable
Delete a conversation, a file, or your whole account. Deletion is real, not a hidden flag.
FAQ
Questions, answered
How does Nitrah decide which model to use?
Every request is inspected for what it actually needs — images, long context, step-by-step reasoning, tools — and matched against a registry of models with their capabilities, context limits and cost. The best fit for your plan wins. You can always pin a specific model instead, and Nitrah shows you which one answered.
What happens if a provider goes down?
Nitrah fails over to a compatible model automatically and tells you it did. Failover only happens before any text has reached you — once an answer has started, switching models mid-sentence would change the answer, so the error surfaces instead.
What is a credit?
Credits meter usage in proportion to what a request actually costs to run. A short question costs a handful; analysing a long document with a premium model costs more. Your balance and consumption are visible in the workspace at all times.
Do you train on my data?
No. Your conversations and files are used to serve your requests and nothing else. Content is sent to the AI provider that answers a given request, under that provider’s API terms, and is not used to train Nitrah.
Can I delete my data?
Yes. Conversations and files can be deleted individually, and deleting your account removes your workspace and its contents. Usage records are retained in aggregate for billing and fraud prevention.
Can my team share a workspace?
The Business plan adds seats, a shared workspace, admin controls and pooled billing. Every piece of data in Nitrah belongs to a workspace, and access is checked against your membership on every request.