Middle Mann  /  Work  /  AI Workspace

One workspace. Every model. Yours to keep.

A self-hosted multi-model AI cockpit. OpenAI, Anthropic, Grok and any other provider you want to plug in, all reachable from a single interface. Organize your work into hierarchical projects, reuse your prompts and snippets across threads, projects and workspaces, watch your cost in real time and run the whole thing on infrastructure you own outright.

Role Designed and built end to end
Type Self-hosted AI platform
Stack Next.js, TypeScript, Postgres
Because

Context sprawl is the new tab hell.

  • Projects & Workspaces
  • No Lock-In
  • Real Cost Tracking
  • Multi-Model
  • Pin, Reuse, Share
  • Owned Infra
  • Thread Continuity
  • Nothing Lost
  • Compression-Proof
  • Universal Search

Serious work with AI is no longer one chat in one browser tab. It is OpenAI for one task, Anthropic for another, Grok for a third, plus whatever new model dropped this week. Conversations scatter across vendors. Prompts get retyped. Costs are invisible until the bill arrives. Data sits on someone else's servers under terms that change without notice.

The AI Cockpit consolidates that surface into one owned environment. One login. Every provider. Every thread organized into projects and folders you can navigate. The potent moments in any conversation pinned in a click, easy to reference, reuse and share. Every dollar accounted for in real time. Every byte running on infrastructure you control.

The Build

A cockpit, not a chat window.

AI Workspace pairs a modern Next.js interface with a Node and TypeScript backend, Postgres for relational data, Redis for sessions and cache, MinIO for S3-compatible file storage and Server-Sent Events for token-by-token streaming. The result is a workspace that behaves like a real product, not a wrapper around someone else's API.

01Chat. Multi-model conversation with mid-thread model switching over SSE streaming
02Structure. Hierarchical projects and folders nested as deep as the work demands
03Reuse. Saved snippets and clipboard applied across every thread in a workspace
04Cost. Real-time spend telemetry per model, per chat, per workspace
05Files. Owned MinIO storage with presigned URL uploads and previews
06Teams. Multi-tenant workspaces, members, roles and per-workspace billing
The Interface

Designed for professionals and power users.

Project trees, threaded conversations, snippet panels, clipboard, model switching, search across every conversation ever held.

Built by operators for operators. Nothing decorative. Nothing hypothetical.

UI walkthrough / silent demo
  • Syntax Highlighting
  • Inline Edit
  • Live Token Count
  • Folder trees
  • Markdown Preview
  • Drag-Reorder
  • Resizable Panes
  • Message Jump
  • Threaded search
  • Model presets
  • Keyboard-first
  • Thread Switcher
The Controls
Settings walkthrough / silent demo

Every knob exposed.

A platform you own is a platform you should be able to tune. The settings layer is deep on purpose. Account, personalization, notifications, team management, integrations, API keys, billing, usage. Nothing hidden behind a paywall or a support ticket.

  • Spend caps & alerts
  • Team Roles
  • Per-Space Billing
  • Data Export
  • Session Timeouts
  • BYO API Keys
  • API & Connectors
  • Notifications
  • Model Overrides
  • System Prompts
  • Per-Model Usage
  • Personalization
The Stack

Modern. Boring. On purpose.

Every choice on this build optimizes for ownership, longevity and developer sanity. Modern enough to move fast. Boring enough to still run in five years without a rewrite.

Frontend
Next.js 14 React 19 TypeScript Tailwind 4 SWR dnd-kit react-markdown recharts
Backend
Node.js Express 5 TypeScript Prisma ORM JWT Auth Zod Pino Logs SSE Streaming
Infrastructure
PostgreSQL Redis MinIO S3 Nginx Let's Encrypt Self-Hosted
AI Providers
OpenAI Anthropic Grok BYO Any Provider Per-Model Telemetry
FAQ

Frequently asked questions.

What is a self-hosted AI workspace?

A self-hosted AI workspace is a multi-model AI interface that runs on infrastructure you own. Instead of routing your conversations through a vendor's servers, the frontend, backend, database and file storage all live on your own hardware. You bring your own API keys for OpenAI, Anthropic, Grok or any other provider, and your prompts, threads and files never leave your control.

How is AI Workspace different from ChatGPT or Claude.ai?

ChatGPT and Claude.ai are single-vendor interfaces owned by the model provider. AI Workspace is a multi-model cockpit you own outright. Conversations, projects, snippets and cost telemetry span every provider you plug in. You can switch models mid-thread, organize work into hierarchical projects and folders, reuse snippets across every conversation, and see real-time spend per model. Nothing routes through anyone's terms of service except the model providers you explicitly call.

Which AI providers does AI Workspace support?

OpenAI, Anthropic and Grok are supported out of the box. The BYO API Provider architecture accepts any OpenAI-compatible endpoint, which covers most major providers plus local models via Ollama or LM Studio. New providers are a configuration entry, not a code change.

Do you keep my prompts, threads or API keys?

No. Because AI Workspace is self-hosted, every prompt, thread, snippet, uploaded file and API key stays on your server in your Postgres database and your MinIO storage. Middle Mann has no visibility into any of it. There is no shared inbox and no vendor telemetry.

What is the technology stack?

Frontend is Next.js 14, React 19, TypeScript and Tailwind 4. Backend is Node.js, Express 5, TypeScript, Prisma ORM and JWT auth. Infrastructure is PostgreSQL, Redis, MinIO S3-compatible storage and Nginx behind Let's Encrypt TLS. Streaming is Server-Sent Events. Deployment is single-server or clustered on hardware you control.

Can multiple team members share the same workspace?

Yes. AI Workspace is multi-tenant with workspaces, members, roles and per-workspace billing. Teams share projects, snippets and files while cost telemetry stays scoped per user and per workspace.

How much does AI Workspace cost?

Middle Mann builds AI Workspace as a bespoke engagement, not a SaaS subscription. Cost depends on scope, integrations and deployment complexity. Contact Middle Mann for a scoped quote.

The Outcome

Software Sovereignty.

The code, the data, the models, the hardware it sits on, all of it belongs to you. Nothing routes through a vendor's terms of service. Nothing gets deprecated by a quarterly roadmap. Every panel can be reshaped, every setting overridden, every workflow extended to fit the way you actually work.

Models, one interface
0
Cloud vendor dependencies
100%
Data and infrastructure owned
Build yours

Want a workspace like this for your operation?