One substrate.
Five layers.

Start with a chat window. End with assistants running the processes your business depends on, unattended. Every layer sits on the same context, the same permissions and the same provenance record.

Against the workbenches

Use the best model. Do not bet the building on it.

ChatGPT Enterprise, Claude for Work, Gemini and Copilot are strong at what they do. They are also single-vendor. You pick one provider and build on their models, their terms, their jurisdiction. SynthWise is LLM-agnostic. OpenAI, Anthropic, Google, Mistral, IBM Granite or a model on your own iron, through one interface. Switch when the market moves. We often use those models through our adapters. We position against lock-in and data control, not against the models.

Against the automators

Built for AI processes, not adapted from connectors.

n8n, Make, Zapier and Power Automate are excellent at connecting SaaS applications. AI was added later, usually as a step in an integration-centric workflow. SynthWise was built with AI as the core: knowledge, context, memory and multi-model routing are first-class. We will not match their connector counts. We compete on what happens between the integrations, in regulated environments.

LAYER 01 - START HERE

Chat & prompts

Multi-model chat. Send one prompt to several models side by side.

Keep the prompts that work. See which model earns its place. No lock-in from day one, because the conversation is not bound to a single vendor's window.

The workbench products optimise for their own model. This layer exists so you can compare, then keep the winner without rebuilding the rest of the stack.

LAYER 02 - ROLES

Assistants

Purpose-built AI roles with their own instructions, tools and project access.

An ops assistant, a credit-review assistant, a delivery lead, each one bounded, each one auditable. Give one a workflow to run and it acts as an agent, taking multi-step action within the same guardrails.

A chatbot with a persona is not a role. A role has tools, permissions, a project, and a record of what it was allowed to touch.

LAYER 03 - KNOWLEDGE

Context

Workspaces, projects, files, schemas, memory and retrieval.

The business knowledge that makes a model useful instead of merely fluent, maintained by you, on your terms. When an asset is included in a prompt, that inclusion is logged.

Most platforms treat files as attachments to a chat. Here they are first-class assets, with sensitivity and an external-sharing flag, because that is how you keep a DPO on side.

LAYER 04 - AUTOMATION

Workflows

Triggers, nodes, conditions and human gates, wired into systems with a REST or webhook API.

Xero, Slack, HubSpot, Trello and Google Calendar out of the box, and many more. Quote-to-invoice, month-end reporting, customer onboarding, supplier chasing, approvals, specified once, run forever.

The human gate is a node, not a hope. A run can pause, wait for a named person, then resume. That decision is a record, not a comment in Slack.

LAYER 05 - BUILD

Studios

Prototype, design and ship inside the platform itself.

Create images or video. A Prototype Studio session becomes a Design Studio spec becomes a Coding Studio pull request, one thread, one provenance record, from idea to merged code.

The consultancy is the platform. The thing we use to build your software is the same substrate that will run it.

What it is built on

Rails 8. PostgreSQL.
You can read it.

The stack is Ruby on Rails 8 and PostgreSQL, with vector search where retrieval needs it, Solid Queue for jobs, no Redis tax. Self-hostable. We name it so a CTO can inspect the architecture. If you leave, you take your data, context and workflows with you, in standard formats. The platform itself stays with us.

Next step

Bring the process
everybody dreads.

We map it, run it on SynthWise, and hand you the receipts. If it works, the next process is mostly configuration.

One substrate.Five layers.