AI · Context orchestration

Take back control of AI context.

Agnis’s Context Waterfall puts you in charge of exactly what reaches the model, so you shape the answer instead of hoping for it, get it right in one go, and stop burning time and credits on bad outputs.

Get early access (Free)See how it works

No card required for the trial.

You’re paying the context tax.

Every session starts by priming the AI. Your background, your rules, the state of the project, typed in again, or assumed known with no readout of what it actually has. Then you hand it over and hope. When the answer comes back wrong, you can’t tell which part caused it.

Two problems wearing one coat. The context tax: hours lost re-priming AI, and reworking answers it built on irrelevant context. AI sprawl: productivity leaking into the gaps between tools that don’t talk to each other.

How it works today

  • Your company background, your title and your rules, pasted in again, every time.
  • No say in which context gets used. One global instruction applied to everything you do.
  • Enormous context windows stuffed with far too much, none of it tailored to the task.
  • Google Workspace for one thing, Claude for another, a company tool for a third. Nothing joins up.
  • A weak answer, three more attempts to fix it, and credits gone on all four.

How it works with Agnis

  • Say it once, at the level where it’s true. Everything below inherits it.
  • See every piece of context going with your request, and switch any of it off.
  • Precise, tailored context instead of a bloated window. Sharper answers, fewer credits.
  • Every major model in one workspace. Switch mid-conversation without losing the thread.
  • Work lives in a tree you can navigate, not a list you scroll.
Work a different way with AgnisFree while we’re in early access. No card required.

The Context Waterfall

The mechanism the rest of it rests on. Context flows downhill: anything true higher up is automatically true for everything beneath it, and at every level you hold the tap. You decide what flows through and what doesn’t.

1

Lay the foundations

Who you are, what your business does, what your role is. Agnis interviews you. A short conversation, not a form. These foundations are inherited everywhere, by everything, forever.

2

Build layers on top

A client, a project, a launch. Each layer adds its own context to everything it already inherited, and layers nest inside layers, as deep as your work actually goes.

3

Control what’s sent

Every request shows you exactly which pieces of context are travelling with it. Switch any of them off for this task without deleting anything. This is the part no other tool gives you.

4

Get it right first time

Precise context in, strong answer out, instead of three rounds of correction. Better outputs and less spend, from the same models everyone else is using.

And then you decide what actually goes.

Inheritance without control is just a bigger prompt. Every request shows you the exact pieces of context travelling with it, and each one has a switch.

Turn the folder brief off for this one post because it isn’t relevant. Nothing is deleted. It stays in place for everything else, and comes back next time. This is the part no other tool gives you.

Build your first layer

A workspace, not a chat box.

Documents that write themselves

Talk in the chat; the document updates in the editor beside it. Edit it yourself at any point. Agnis notices and works around your changes.

Emails and posts, in your voice

Purpose-built types for the things you actually send. They inherit the same tone rules you set once, in Profile, on day one.

Live web search

Agnis searches the web when a question needs current information, and tells you when it's doing it. Works across every provider.

Drop in your files

PDFs, Word, PowerPoint, spreadsheets, images. Attach them to a layer so everything underneath can draw on them, or to a single message.

Talk instead of typing

Dictate straight into any conversation. Useful when you're thinking out loud, which is usually when the context worth keeping comes out.

Your keys, your data

Add your own provider keys and pay the model providers directly at their rates. Keys are encrypted at rest and never readable from the browser.

Every major model. One workspace.

No separate accounts, no comparing tabs. Switch model in the middle of a conversation and it picks up exactly where the last one left off. The history is the history, whoever wrote it.

AnthropicClaude
  • Fable 5
  • Opus 4.8
  • Sonnet 5 (default)
  • + 5 earlier Claude models
OpenAIGPT
  • GPT-5.6 Sol
  • GPT-5.6 Terra
  • GPT-5.6 Luna
  • + 5 earlier GPT models
GoogleGemini
  • Gemini 3.6 Flash
  • Gemini 3.5 Flash
  • + 2 Flash-Lite models
MoonshotKimi
  • Kimi K3
  • Kimi K2.7 Code
  • + Kimi K2.6

Don’t know which model to pick? Don’t pick one.

Leave it on Auto and Agnis chooses for each message: a fast, cheap model for a quick question, a heavyweight for a hard piece of reasoning, and a search-capable one when your question needs today’s information. You can always override it. New models are added as providers ship them, so you don’t have to track releases.

Professionals shouldn’t adapt to AI. AI should adapt to professionals.

Most people who bounce off AI didn’t get a bad model. They got a blank box and no idea what to put in it. Agnis removes the part you were being asked to be good at, and three shifts sit underneath every decision we make about it.

Power tool
Guided workspace
Agnis can tell where its picture of your work is thin, and asks you about it. You are never left guessing what it doesn’t know.
Output
Outcomes
Success isn’t “the AI wrote something”. It’s that the document is finished, the goal is met, and the intent you started with didn’t drift along the way.
Chat
Continuity
Chat is ephemeral. Work isn’t. Your context, decisions and drafts persist as structure you can return to, not a transcript you have to scroll.

You’re the architect. Agnis is the engine.

This isn’t a replacement workforce. It’s a power suit. You provide the direction and the judgement, the parts that are actually yours, and Agnis does the heavy lifting of remembering, routing and executing. The human stays in the driver’s seat, which is the only arrangement that produces work you’d put your name on.

You shouldn’t need to learn prompt engineering to get a good answer.

Agnis asks you questions rather than waiting for instructions. It speaks first when you open a new piece of work. You can dictate rather than type, which is when the useful context tends to come out anyway. And it writes and keeps the underlying context itself, so the thing that separates a mediocre answer from a good one is handled for you, permanently, instead of being retyped badly each time.

You never deal with
You do this instead
Writing a system prompt
Answer three questions once
Choosing a model
Leave it on Auto
Tokens and context windows
Nothing, it’s handled
Pasting your background in
It’s already there

The Agnis Engine

Context is the foundation, not the finished building. The end state is a single flow where one instruction moves through research, analysis, drafting and delivery without you changing tabs, because Agnis is holding the context the whole way through.

Live now

Context Manager

The Context Waterfall. Remembers everything about you and your business, inherits it down the tree, and lets you control exactly what travels with each request.

No more repetitive priming
Live now

Model Router

Claude, GPT, Gemini and Kimi behind one interface. Agnis picks the right one for each message, or you choose. New models appear as providers ship them.

One workspace, not four tabs
Planned

Integration Layer

Connections out to the tools your work already lives in, starting with the ones people ask for most. The context you’ve built becomes available to them, rather than being trapped here.

Everything happens in one place
Planned

Single Flow

Chaining steps together: research, then analysis, then draft, then send, as one instruction rather than ten. The Idea Accelerator: five words in, a finished piece of execution out.

Ten steps become one
Sign up and shape itEarly users have a direct say in what gets built next.

Why this exists.

Andrew Whitbread, founder of Agnis
“I kept getting outputs that weren’t good enough and I couldn’t tell you why. I had no visibility of what the model was actually working from, and no way to change it. That’s not a model problem. That’s a platform problem.”

I’m a Head of Product in financial services, working on payments and digital assets. Everywhere I’ve worked has the same setup: three or four AI tools, none of them talking to each other, everyone rebuilding the same context by hand in whichever tab they happen to be in. It’s slow, it’s inconsistent, and it quietly costs a fortune in retries.

So I built what I wanted to use. One workspace, every model, and a structure that shows me what each request is built from and lets me change it before I send it.

It’s early. I’d rather have fifty people using it properly and telling me what’s wrong than five thousand signing up and drifting. If you get in now you’ll talk to me directly, and what you say will change what gets built.

Get early access
Andrew WhitbreadFounder, Agnis · LinkedIn · andrew@agnis.io
Mission

To equip the modern workforce with a guided, contextual engine that unifies AI tools into a single flow, so professionals can think clearly and accelerate their ideas without the friction of prompt engineering.

Vision

To become the primary business interface for the AI-first era: the standard for how knowledge turns into action.

Free now. £5 a month for life if you join in early access.

No trial countdown, no card on file, no surprise invoice. When pricing starts you’ll be told well before it applies to you, and everyone in early access gets notice, not a bill.

Available now

Early access

While we’re in beta

Free

No card required

  • The whole product, nothing held back
  • A funded onboarding journey to get you started
  • Add your own key at any point to carry on without limits
  • Locks in £5 a month for life when paid plans begin
  • Direct line to the founder
Get early access

Standard

Bring your own key · from launch

£10 / month

£5 / month for life if you joined in early access

  • The full workspace: waterfall, every item type, every model
  • Plug in Anthropic, OpenAI, Google or Moonshot keys
  • Model usage billed by the providers at their own rates, nothing added by us
  • Spend caps and per-provider budgets built in
  • Cancel whenever, and your context stays yours
Join now, lock the price

Pay as you go

Planned, no keys needed

Usage + 5%
  • We handle the provider accounts, so you never touch an API key
  • Pay for what you actually use, with no monthly commitment
  • Spend $100 on models, pay $105, and see the breakdown
Start free today

Why we think this pays for itselfWhat we believe

We believe better context means fewer messages, less rework, and a faster route to the outcome you actually wanted. Whichever way you pay, the models are the same models at the same rates. What changes is how many attempts it takes to get there, and over time that is where the value sits.

Three steps and you’re working.

Create your account (Free)Already have one? Sign in

The things people ask.

How is this different from ChatGPT or Claude?
Those tools start each conversation from nothing, or from one global instruction that has to serve every kind of work you do. Agnis stores your context as a tree. Anything you write higher up is inherited automatically by everything beneath it, so a chat about next quarter's launch already knows your business, your role, your tone and the launch, without you doing anything. You can also use Claude, GPT, Gemini and Kimi from the same place, which no single provider will offer you.
Do I need my own API key?
Not to start. Early access includes a funded onboarding journey so you can try the product properly without one. The paid plan is bring-your-own-key: £10 a month for the workspace, or £5 a month for life if you joined during early access, and you pay Anthropic, OpenAI, Google or Moonshot directly for what you use, at their rates, with nothing added by us. If you'd rather not deal with keys at all, the planned pay-as-you-go option covers that. Keys are encrypted at rest and are never readable from the browser.
What's actually built, and what's still coming?
Live today: the Context Waterfall with per-task control over what's sent, chats, documents, emails and blog posts, file uploads, web search, and every model across Anthropic, OpenAI, Google and Moonshot with automatic routing. Still to come: connections out to the other tools your work lives in, and chaining several steps into one instruction. The roadmap section above labels which is which, and we'll keep it honest as things ship.
What happens to my data?
Your context, chats and documents are yours. They're stored in the EU and isolated per account at the database level. Self-serve export is on the roadmap, and until it ships we'll provide a copy of your data on request. We don't train models on your content. We don't train models at all. Model providers process your requests under their own API terms, which for the providers we use means no training on API traffic.
I'm not technical. Is this for me?
Particularly. Agnis asks you questions instead of waiting for instructions, picks a suitable model for each request, and writes the reusable context in the background. You don't need to know what a token, a system prompt or a context window is, and nothing on screen will ask you to.
What will it cost, and when?
Free during early access. From launch it's £10 a month with your own API keys, and you pay the model providers directly at their published rates with nothing added by us. If you join during early access, your price is £5 a month for as long as you keep your account, not £10. A pay-as-you-go option is planned for people who'd rather not manage keys, priced at around 5% above usage. Nobody gets charged without clear notice first.
How does this save me money if the models cost the same?
Because you spend less getting to the answer. Most of what people pay for isn't the good output. It's the two or three attempts before it. When the model starts with precise, relevant, current context that you've chosen, the first answer is far more likely to be the one you keep. That's the saving, and it compounds the more you use it.
Can I switch models halfway through something?
Yes, and it's designed for it. The conversation history is the conversation history regardless of who wrote each turn, so you can start on a fast cheap model, hit something hard, move up to a frontier model for one answer, and drop back down. Leaving it on Auto does this for you.
What if you shut down?
Fair question for anything this early. Everything you write is stored as plain Markdown, so there's no proprietary format holding it hostage. Self-serve export is on the roadmap; in the meantime ask us and we'll send you a copy of your data. And if you're on your own API keys, your relationship with the model providers doesn't run through us at all.

Take back control.

Early access is open and free. Sign up, build your context once, and every piece of work from then on inherits it.

Get early accessSee how it works

No card required. Cancel at any time.