Jump to01Why a workspace02Building blocks03Admin controls04Safety & governance05By role06How it fits07FAQ
The OS Workbench

Your team's private AI workspace.

Catalyst goes beyond planning. The OS Workbench gives your team a private AI workspace — chat, skills, scheduled automations, brand-styled exports — all under your own LLM key, governed by your policies, and audited end-to-end.

Start 14-day trial See the building blocks →
01 · Why a workspace

A team can't run AI on personal accounts.

Most companies' AI usage is a sprawl of individual subscriptions and browser sessions. It works — sort of — until you want to share what's working, govern what's allowed, or know what it's costing. Three problems every team eventually hits:

“Our team uses personal ChatGPT accounts.”
A great prompt one person writes never reaches the rest of the team. Knowledge lives in someone’s browser history. Spend is invisible. There’s no audit trail when someone leaves.
“Generic AI tools don’t know our business.”
ChatGPT and Claude.ai don’t know your tools, your customers, your policies, or your tone. Every conversation starts from scratch. Skilled prompters do well; everyone else gets generic output.
“IT can’t tell what’s happening with AI.”
No record of which tools touched which data. No way to enforce policy. No way to prove compliance to insurers or auditors. AI use is shadow IT, by definition.
02 · The building blocks

Nine surfaces. One workspace.

Each surface is something a teammate uses; together they replace the patchwork of personal AI subscriptions with one governed system that compounds with use — and the accumulated preferences, prompts, and response style stay portable so the value never gets locked in the vendor.

01Chat
Strategic question
What if every conversation already knew our business?

Streaming chat with Claude — your company’s context, built in.

A real Claude session, in a team workspace. Streaming responses, web search, image uploads, drag-drop documents, and one-click “Save as result” — then share it with specific teammates.

What it changes

Conversations are persisted, searchable, and grouped by day. The org’s conformance policy auto-attaches to every session. Switch models inline. Invoke a Skill mid-thread with /.

Business impact
  • Web-aware: Claude can search the live web for current info — no more “my data ends in 2024.”
  • Multi-modal: Paste screenshots or upload images; Vision handles charts, error messages, photos.
  • Audit-friendly: Every message is persisted and tied to a user, model, and cost.
02Skills
Strategic question
What if a great prompt could run itself — by anyone, every time?

Reusable AI skills with typed inputs and structured outputs.

A skill is a saved prompt template with named inputs and a target output format (Markdown, PDF, DOCX, PPTX, JSON). Build it once; anyone on the team runs it on demand.

What it changes

Skills can scope inputs (e.g., “Company URL”, “Target audience”), reference attached documents, and auto-generate a saved result when finished. Permissions: admin-created vs. user-created visibility rules.

Business impact
  • Distribution: The best prompter on your team becomes a force-multiplier for everyone else.
  • Consistency: Same inputs → same structure of output, every run.
  • Auditable: Every run captures inputs, output, cost, and model used.
03Documents
Strategic question
What if Claude already read every page of our handbook?

Drop files in once. Search them, fold them, reference them everywhere.

Drag in PDFs, Word docs, or text — up to 100 MB each, dozens at a time (parallel uploads with live progress). The platform extracts and indexes them for semantic search, then makes them attachable as context for chat and skills.

What it changes

Organize documents into Folders and attach a whole folder to a skill, sub-chat, or chat — it pulls in every document in that folder, including ones you add later. Semantic Document Search finds the right passage by meaning and opens the source PDF to the exact page. Admin-scoped uploads auto-inject as “Organization context” on every chat.

Business impact
  • Find it by meaning: Search “refund window” and get the exact clause — even if the doc never uses that word.
  • Organize once, reuse everywhere: Group manuals into a folder; attach the folder, not 40 files.
  • Single source of truth: Update the doc, every future chat and skill updates with it.
04Saved Results + Exports
Strategic question
What if every chat could become a polished deliverable?

Click any AI output. Export it as a brand-styled PDF, Word doc, or PowerPoint.

Every Skill output and any chat message you save becomes a versioned saved result — private to you, shareable with teammates. From the result page, export as PDF / DOCX / PPTX in two modes: Quick (deterministic, free, ~2-5s) or Smart (Claude restructures the content for the target medium first, ~$0.02-0.10).

What it changes

Smart export accepts custom instructions — “make this a 5-slide deck for execs,” “formal report tone,” “add an executive summary.” Re-render any time with new instructions.

Business impact
  • No more reformatting: Stop pasting chat output into Word and fixing it for an hour.
  • Format-aware: A presentation comes out as slides; a report comes out with a cover page and ToC.
  • Versioned: Every export is preserved; re-roll without losing the original.
05Routines
Strategic question
What if your competitive intel scan ran itself, every Monday?

Schedule any skill on a cron. Get the result in your inbox.

Turn any Skill into a Routine that runs on a schedule — weekly, monthly, business-day-only. Cloud Scheduler runs it, emails the result, and keeps a run history.

What it changes

Timezone-aware, enable/disable per routine, automatic failure backoff, notification preferences per recipient.

Business impact
  • Reliable cadence: Reports that used to be “I’ll do it Monday” become “it’s in your inbox Monday.”
  • Compound value: A skill written once runs hundreds of times — the ROI of one prompt compounds.
  • Hands-off: You set it up; the workbench runs it.
06Company Prompts
Strategic question
Where does our team’s prompt wisdom live?

A team prompt library that doesn’t live in Notion.

Save prompts your team actually uses. Search, pin, copy. Difficulty tagging so people pick prompts that match their experience level.

What it changes

Different from Skills (which run inside Catalyst); these are copy-and-paste prompts that go anywhere — Claude.ai, ChatGPT, internal tools.

Business impact
  • Knowledge stays: When someone leaves, their best prompts don’t leave with them.
  • Onboarding wins: New hires get the team’s collective prompt know-how on day one.
07Advisor
Strategic question
When the AI can’t answer it, who can?

One click to a real human — with all your context already loaded.

Book a NevTech advisory session from inside the workbench. Strategy session, implementation sprint, or fractional AI officer. The booking carries your context so the call starts solving, not introducing.

What it changes

Three tiers, flat-fee, no agency retainers. Same advisor mesh that runs the methodology side of Catalyst.

Business impact
  • Skip the discovery call: Advisors already see your assessment, roadmap, and recent runs.
  • On-demand: When you need a human, you don’t wait two weeks for one.
08Sub-Chats
Strategic question
What if you could hand a customer a specialist that only knows what it should?

Specialized, shareable assistants scoped to the docs you choose.

Spin up a focused assistant with its own instructions and a scoped set of documents (or a whole folder). Keep it internal for your team, or publish a shareable link — open or password-protected — so customers and field techs can use it with no login.

What it changes

Public links never expose your admin docs, tools, or other context — only the sub-chat’s chosen documents. Per-conversation limits and rate limits keep costs in check; all usage bills to your existing key.

Business impact
  • A product, not a prompt: Ship a “Tech Support” or “Product Q&A” bot in minutes, branded and scoped.
  • Safe to share: Outsiders get answers from exactly the docs you picked — nothing else.
  • Always current: Point it at a folder; new manuals you add show up automatically.
09Your Profile (Soul) + Company MD
Strategic question
What if every AI answer already matched your voice — and travelled with you if you ever leave?

Portable per-person and per-company response profiles. Your accumulated AI IP, in a plain markdown file.

Your Soul is a personal markdown file describing how you want AI to respond — tone, structure, length, do's and don'ts. Your organization gets a Company MD that codifies house style for everyone. Both auto-inject into every chat and skill run so answers match your voice from message one, with no re-teaching each session.

What it changes

Both are downloadable as .md files at any time. Paste them into ChatGPT, Claude.ai, or any other LLM tool and get the same behavior elsewhere — the profile is your IP, not the vendor's. Company MD overrides personal Soul on conflict (spelled out in the wrapper text the model actually sees), and both are scoped strictly to style/tone so Legal Mode's grounding and citation rules always take precedence. Version history means every save is one-click revertible.

Business impact
  • No vendor lock-in: Your accumulated response preferences leave with you if you ever leave Catalyst — a real switching cost only if we earn it every day.
  • House style, enforced: Company MD makes every attorney or CS rep sound like your firm from message one — junior team members get partner-quality format automatically.
  • Answers to procurement: "What happens to our AI IP if this vendor gets acquired or shuts down?" — you own it, always, as a plain markdown file.
03 · Admin controls

Built for the person who has to explain it to the board.

Six admin-only surfaces give operators and IT the controls they need to deploy AI without losing sleep.

Members
Add team members manually with email + password. Assign roles. Remove on offboarding.
LLM Keys
Bring your own Anthropic / OpenAI / Gemini / Azure key. Per-key daily + monthly spend caps.
Connections
OAuth into Gmail, Calendar, Drive, QuickBooks, HubSpot — with revocable tokens.
Analytics
Cost, tokens, and usage rollups by skill, user, and model. 7/30/90-day windows.
Audit Log
Immutable trail of every state change: role changes, key rotations, exports, deletions.
Settings
Org-wide chat system prompt. Admin documents. Auto-attach conformance policy to every chat.
04 · Safety & governance

The opposite of shadow IT.

Personal ChatGPT accounts can’t be governed. The Workbench can. Every piece of usage is keyed, capped, encrypted, and audited.

BYOK
Your team’s usage runs on your own Anthropic / OpenAI / Gemini key. NevTech never touches your data via our key.
GCP KMS envelope encryption
Keys are envelope-encrypted at rest — a data key wraps each API key, encrypted again by GCP KMS. Plaintext never logged.
Spend caps
Per-key daily and monthly limits. The orchestrator refuses calls that would exceed them — no “first we learn there was a $4k bill”.
Per-org row-level security
Postgres RLS enforces tenant isolation at the database layer — not in application code. One org’s data is physically unreadable by another.
Immutable audit log
Every state change — role grant, key rotation, skill delete, result export — is logged with actor, timestamp, and before/after snapshot.
Conformance auto-attach
One admin toggle injects your approved AI Conformance Statement as “## Organization context” on every chat and skill run. Policy enforced, not just published.
05 · By role

What the workbench does for the person looking at it.

Different seats, different value. Pick the one that’s yours:

See every AI dollar your team is spending in one analytics view — by user, skill, and model.
Your AI Conformance Statement auto-attaches to every chat and skill run — policy enforced, not aspirational.
Onboard a new team member in 30 seconds: add them, share the password, they sign in with all your skills and documents ready.
A documented record of every AI conversation — audit-ready for insurers, lenders, or a future acquirer.
06 · How it fits

Planning lives in Catalyst. Execution lives in the Workbench.

The methodology side of Catalyst — Discovery, Roadmap, Project tracking, Conformance — answers what should we do? The Workbench is where your team actually does it. The two stay in sync automatically: your Conformance Statement enforces policy in the Workbench; your Skills produce the deliverables your roadmap calls for; your Workbench usage shows up in your platform analytics.

You can run either side independently — the methodology side without the Workbench, or the Workbench without ever opening the methodology dashboard — but they’re designed to compound when used together.

07 · FAQ

Common questions.

Do we need our own Anthropic key?+
Yes — the Workbench is BYOK. You supply an Anthropic, OpenAI, Gemini, or Azure key from your own provider account, and your team’s usage bills to you directly at provider rates. NevTech never runs your team’s requests through a NevTech-owned key.
What does a Smart export cost?+
Typically $0.02–$0.10 per export, billed to your BYOK key. The cost depends on the source length and the target format. Quick exports are free — they’re deterministic renders, no LLM call.
Who can see what across the team?+
Admins (OWNER / ADMIN roles) see everything in the org — every chat, skill, document, saved result, and analytics view. Members see their own chats and the org’s shared skills, documents, and Company Prompts. Saved Results are private to whoever created them — shareable, view + download, with specific teammates. Viewers get read-only access to their own saved results and chat history.
Can we cap spending?+
Yes — every LLM key has independent daily and monthly spend caps. The orchestrator refuses calls that would exceed them and surfaces a clean error to the user, with a link for the admin to raise the cap if they want to.
Can we white-label it for our customers?+
Not in v1. White-labeling is on the roadmap; talk to us if it’s a dealbreaker and we’ll prioritize it.
What happens to our data when we cancel?+
Documents, chats, saved results, and audit logs are exportable as a single archive any time. After cancellation we retain the data for 30 days (in case of accidental cancellation) then permanently delete it. Anthropic / OpenAI data retention is governed by your provider agreement — not by NevTech.
Try the OS Workbench

Bring your team’s AI
out of the browser.

14-day trial. From $49/mo. Your BYOK Anthropic key, your data, your policy — in a workspace your team can actually share.

Start 14-day trial See pricing