Product · Foundation

Every model your firm uses, through one gateway you control.

The Gateway plugs into the frontier models your firm already uses, Claude, GPT and more, alongside our own finance-native model. It redacts what should never leave, keeps every answer tied to your sources, and puts cost, access and audit under one set of rules. Studio, Desktop and the Orchid add-on run on it, and your engineers point their own systems at it with one configuration change.

What the gateway does

Plugs into the frontier models you use

Claude, GPT and the other leading models, on your firm's own agreements. Nothing to replace and no new vendor to approve: the gateway sits in front of what you already have.

The right model for every request

Our inference engine sends each request to the model that does it best for the least. You choose how much quality to trade for cost, or none at all.

Sensitive data stays inside

A redaction engine replaces client names, personal details and the terms your firm defines before a request reaches any outside model, and restores them in the answer. Your admins set the policy; nobody can loosen it.

One source of truth

Every model draws on the same approved documents and data, and answers can be checked against their sources before they reach your team, with anything unsupported flagged.

Cost you can see and steer

Spend by team and by person, the saving against what you would have paid, and settings that keep it where you want it.

One set of rules for every model

Retention, access and audit are set once for the firm and apply to every model and every product, whoever built it.

A complete record

Every request, answer and source can be written to a store your firm owns, so the audit trail is yours whichever model did the work.

Your data services, everywhere

Connect FactSet, LSEG, S&P and your document stores once, with your firm's own licences, and they are available in every Orchid product.

Our own model, included

orchid01, our finance-native model, sits alongside the others for the work where it is the strongest choice.

Our own model

orchid01

A finance-native model, trained by us on financial reasoning we generated and graded ourselves, and available through the gateway alongside every other model. It is not a wrapper around a general model with a longer prompt in front of it.

Shipping

What it was trained to do differently

#BehaviourThe mistake it prevents
01Never retypes a numberA figure copied from memory instead of from the data
02Checks before it claimsA conclusion stated first and verified afterwards, or never
03Says how sure it isOverconfidence, and hedging so heavy the answer is useless
04Holds its positionFolding the moment someone disagrees without new evidence
05Reads the whole documentMissing the fact that only appears when two exhibits are read together
06Gets the mechanics rightMillions read as billions, dropped negative signs, invented line items
07Declines the unknowableAnswering confidently about a period it has no information on

Each of these is measured, not asserted. We will share the full evaluation set with your team under NDA.

For your engineers

One line changes. Nothing else does.

The endpoint speaks the same shapes as the SDKs your team already writes against, including streaming and tool calling, so existing internal tooling and agent loops run unchanged.

Grounding, on request

Ask for it and a second pass checks the answer against the documents supplied, returns a score, and marks anything it could not verify, or rewrites it.

Documents as tables

Financial PDFs are turned into structured tables the model cites from, rather than prose it has to re-read.

Your own audit record

Every request, response and source can be posted, signed, to an endpoint you own, so you keep a complete record even when we retain nothing.

The migration

01# change the base URL, keep everything else
02client = OpenAI(
03 base_url="https://llm.orchid.ac/v1",
04 api_key=os.environ["ORCHID_API_KEY"],
05)
06 
07resp = client.chat.completions.create(
08 model="orchid-auto",
09 messages=messages,
10)

Keep your key management, your prompts and your existing loops.

On the same endpoint

Orchid for Excel

An add-in in the workbook your team already lives in. It reads the sheet, writes cells and formulas, and traces what depends on what, so the output is a model an associate can audit rather than a paragraph describing one.

Shipping

It talks to your firm's endpoint, under the same access and retention settings as everything else.

Deployment options

Run the whole thing yourself

For firms that want nothing on our side: conversations, files, indexed documents, scheduled work, dashboards and logs all move to a database your firm owns and runs. We hold sign-in and usage counts, and no financial content at all.

Shipping

Integrates into

Drops into the stack you already run.

No migration, no rewrite, no second set of credentials to manage.

Existing SDKs

The same request shapes your team already writes against, streaming and tools included.

Coding agents and editors

Anything that accepts a custom endpoint works by changing the base URL.

Excel

The add-in reaches the same endpoint, under the same rules.

Your frontier models

Claude, GPT and the other leading models, on your firm's own agreements. The Orchid add-on brings the gateway into Claude itself.

Your identity provider

Access follows the accounts and groups your firm already manages.

Your log and audit store

Signed records of every request delivered to infrastructure you own.

Bring your engineers to the call.

Thirty minutes, and they can interrogate the endpoint, the retention settings and where every byte physically sits.