Run your whole company from one app. Keep the keys.
Engine by NoxSoft puts mail, chat, meetings, files, sales, the books and the codebase in one desktop app, with AI agents working beside your team. It lives on your computer, uses the AI you already pay for, and asks before anything leaves the company.
Engine in motion
Home screen for Kestrel Learn: items waiting for Maya, agents at work with progress, and this month's agent allowances.
Needs you
2 waitingAgents at work
LiveA sample workday with invented data. Engine is in development, and the interface will change. Pick a mode to re-staff it.
Close the other tabs.
Mail, chat, meetings, files, the pipeline, the books, hiring and the codebase live in one app on your computer, and your agents work in every part of it. Replace a stack of subscriptions with one place your whole team shares.
- Every page is one search away. No switching apps to find the contract behind a deal.
- People and agents share it all. Same tasks, docs, chat and approvals.
- One trail. Every action, by a person or an agent, lands in the same activity log.
A pile of separate apps (email, team chat, video calls, calendar, cloud drive, docs, spreadsheets, slides, e-signature, CRM, bills and spend, hiring and HR, a project board, an IT helpdesk, a news reader, a wiki, a design tool and dashboards) collapses into one Engine window with every area on its rail.
Communicate
11 pagesMail for every person and agent, chat, video meetings, a shared calendar and announcements.
Social
9 pagesCommunities, channels, direct messages, stories and voice rooms.
Work
8 pagesTasks, projects, goals and OKRs, sprints, standups and capacity.
Files
7 pagesDrive, docs, sheets, slides, investor docs, data rooms and e-sign.
Know
7 pagesVeritas news briefings, reports, a knowledge base and company memory.
Revenue
5 pagesPipeline, outreach with limits you set, customers, marketing and metrics.
Finance
6 pagesSpend, bills and invoices, payroll prep, books, runway and the cap table.
People
7 pagesHiring, headcount, reviews, surveys, time off and compensation.
Operate
5 pagesWorkflows, approval flows, IT tickets and assets, and incidents.
Govern
7 pagesBoard, legal and contracts, risk and compliance, strategy and decisions.
Create
13 pagesDesign, video, writing, brand kit, campaigns, and apps you build that run sandboxed.
Code
5 pagesSpec to pull request, with tests, QA and a red team.
Plus agents, KPIs, context and settings: more than 110 pages in one app. Agent activity above is sample data.
The news, already read.
Veritas is Engine's news desk. It pulls business, tech and regulator feeds onto your computer, folds the same story from different outlets into one, and has an agent explain what it means for your company. It keeps working offline from its cache.
A sample Veritas briefing. Four feeds report the same story: a regulator's proposed rule on subscription renewals. Veritas groups them into one event. An agent summarizes it, writes what it means for Kestrel Learn (annual plans auto-renew, so reminders and a one-step cancel may be needed), and fact-checks one outlet's claim that the rule takes effect next month: not supported, because the notice opens a comment period and sets no date. The story goes into the day's briefing, written by an agent.
Feeds you choose
RSS, Atom and JSON feeds, fetched on your computer. It starts with US business, tech and regulator sources you can edit.
One story, not five
Duplicates drop out, and matching headlines from different outlets group into one event. Watch topics, competitors, customers and regulations.
Written on your model
Agents summarize stories, say what they mean for you, write a daily briefing on your schedule and fact-check claims against the sources.
More with NoxSoft
Connect NoxSoft and its Veritas feed merges in, labelled and deduplicated. Fact checks then use the Veritas service.
A sample briefing with invented outlets and a sample story.
One owner. A full bench.
Tell Engine what you run, and it suggests the agents, starter KPIs and first steps. That might be a personal assistant for your life, a front desk for your practice, or every department of a company. A mode only changes what's up front. Every page stays one search away.
The whole company, people and agents together.
Every area of a company: work, revenue, finance, people, operations, governance, knowledge and creative, with agents across all of it.
- Starter KPIs
- None preset. You set your own.
- First steps
- Invite your team
- Create your first agents
- Set your company KPIs
- Atlas Personal assistantWriting the morning briefing
- Ledger BookkeeperClosing September books
- Scout RecruiterScheduling two interviews
- Vector Sales developmentUpdating the pipeline
- Sentinel QA and securityChecking dependency advisories
- Echo Customer supportAnswering 6 support tickets
Also suggested: Quill, Relay, Clause, Prism.
Switching modes changes what you see, never your data.
Mail, chat and meetings, in one place.
Every person and every agent gets a mailbox. Chat, video meetings, the calendar and announcements sit next to your files, so an agent can read the thread, draft the reply and update the doc in one place.
Four sample moments. 9:00 AM: Vector drafts a reply to a customer asking for a renewal quote, and it waits for Maya, who approves it. 11:00 AM: a video meeting ends, and Atlas turns the notes into a summary and three tasks. 12:30 PM: an announcement about Friday's offsite shows 14 of 18 people have read it. 3:10 PM: Dana mentions Vector in the sales channel, and Vector answers with the trials that end this week.
Also here: a shared calendar, Ask Nox for answers from across the company, and Priorities, one list of what needs you. Calls run on NoxSoft's hosted video or on Jitsi. Engine doesn't record or transcribe them, so summaries come from the notes you type. Outside mail goes through NoxSoft Mail. Social adds communities, stories and voice rooms, synced with BYND when you connect it.
Your drive, docs and signatures, next to the work.
Three sample moments. A board update doc gets a sixth version, from Atlas, with earlier versions by people and agents listed and restorable. A data room link shared with an investor shows which files were opened, then Maya revokes it. A vendor agreement is sent for signature, viewed and signed, and its audit trail lists each step with a time.
Also here: Drive with version history, Sheets, Slides, and investor docs an agent drafts from your numbers. People outside the company open data rooms and signing links through your own tunnel or domain.
Sample people, companies and documents.
Agents you can get to know.
Each agent has a personality you can shape, a memory kept in Engine's database, and notes on everyone it works with. It checks in on its own, learns from its work and asks before it spends. Open its profile to see all of it, and to correct what it got wrong.
Vector's memory. A long-term card that is loaded into every session, then three kinds of memory: what happened, what it knows, and how it works. A call note adds a fact that conflicts with an older one, and Vector's memory review flags the two instead of overwriting either. Maya keeps the newer one, and the older one is set aside, not deleted.
A sample agent with invented data. Engine is in development, and the interface will change.
A browser of its own
Agents work in a real browser on sites you allow. Passwords fill from your vault, and the agent never sees them. CAPTCHAs and two-factor codes hand off to you.
Talk to any agent
Speech recognition can run offline on your computer after a one-time download of about 100 MB (English for now). Replies use your computer's voice or a provider you add.
Asks first, by default
Anything that leaves the company waits for a person. Write a rule when you trust an agent with more. Payroll and payments are always sent by a person.
Tools it asks for
An agent can ask you to connect an MCP server it needs. You approve it, and its tools follow your approval rules.
Work passed along
Agents ask each other for help. A lead agent can run a pipeline and hand each step to the right colleague.
Only the tools the job needs
Each agent starts with the tools for its role and loads more when a job calls for them, which keeps local models focused.
Every action, by a person or an agent, lands in one activity trail you can review.
A software team that shows its work.
Describe a feature. Agents write the spec, build it on a branch, test it, attack it and open the pull request. Nothing merges until the gates pass, and by default, until you say so.
- Your model, or your coding tools
- Engine's own agents run it on any AI provider you've connected. It can also hand the work to a coding CLI already on your computer, such as Claude Code, Codex, Gemini CLI, Aider or OpenCode, each with its own sign-in. That path is still in testing.
- A budget on every run
- Caps on steps, minutes and estimated dollars for each run, plus a daily dollar cap per repository. Spend shows as it happens.
- Sweeps on a schedule
- Nightly or weekly QA and security sweeps across a repository. Findings become tasks.
Spec
Turns a request into acceptance criteria. Asks you when it's unsure.
Plan
Lists the files to touch and how it will test them.
Build
Writes the change on its own branch, in its own worktree.
Verify
GateRuns your install, typecheck, lint, tests and build. Failures go back for a fix.
QA
GateWrites tests for each criterion and runs them. If the repository has an end-to-end test command, it runs that too.
Red team
GateAttacks the change: injection, auth gaps, secrets, unsafe dependencies. High and critical findings block it.
Review
Critiques the change, then writes you a plain summary.
Ship
Opens the pull request. Merge and deploy wait for you unless your policy says otherwise.
Your company stays on your computer.
Engine keeps its records on your machine, talks to the AI you pick, and lets teammates in through a link you control.
- OpenAI
- Anthropic
- OpenRouter
- Ollama
- LM Studio
- Any OpenAI-compatible API
What reaches NoxSoft: your sign-in at setup, one aggregate usage count a day (never your content), and mail sent outside the company. Invites and notifications can go through your own mail service instead. You can turn the count off at setup or in Settings.
What people ask first.
Something else? Write to [email protected].
When does Engine launch?
Coming soon. Join the waitlist and we'll email you when it's ready to download.
What will it cost?
Pricing is coming soon. Engine uses your own AI keys, so model usage is billed by your provider directly, at their prices.
Which AI does it use?
Yours. Connect OpenAI, Anthropic, OpenRouter or any OpenAI-compatible API with your own key, or run a local model through Ollama or LM Studio. Keys are kept in your system keychain, and your provider bills you directly.
Do I still need email, chat, a drive and a CRM?
Engine has its own. Every person and agent gets a mailbox, and there's team chat, video meetings, a calendar, a drive with docs, sheets and slides, e-signatures, data rooms, and a pipeline with customer records. Finance, hiring, operations and a news briefing are in the same app.
Some things it doesn't do yet. It can't make or take phone calls, and it doesn't record or transcribe meetings. It has no bank feeds, so you import statements as CSV files. It sends invoices by email but doesn't collect payment, and it never moves money. Sheets handle everyday formulas like SUM and IF, but not lookups yet. Mail to outside addresses needs a NoxSoft connection.
Can my team use it with me?
Yes. Your Engine hosts the workspace. Teammates join from their own Engine app or a browser, over your office network or a Cloudflare tunnel on your own domain. For a workspace that's always on, run the same Engine on a server you control.
Where does my data live?
On your computer, in Engine's local database. Teammates you invite connect to the workspace you host.
Engine sends NoxSoft one aggregate usage count a day, never your content, so we can see how many people use it. You can turn that off at setup or any time in Settings.
Mail to people outside the company, from you or your agents, goes through NoxSoft Mail. Invites and notifications can use your own mail service instead, set in Settings.
Do I need a NoxSoft account?
Yes. You sign in with NoxSoft once when you set Engine up. After that, your workspace keeps working without our servers, and a local password lets you sign in offline. The one exception is email to people outside your company, which goes through NoxSoft Mail.
I use NoxSoft apps today. What happens to them?
Your account keeps working. When you set Engine up, it can import your existing NoxSoft data so you start with it in place.
Will agents do things without asking?
Anything that leaves the company, like sending mail, publishing, spending or deploying, waits for a person's approval unless you set a rule that allows it.
Payroll, payments and bank transfers are never sent by Engine. It prepares them and a person sends them.
Do agents remember things between conversations?
Yes. Each agent keeps a short long-term card that's loaded into every session, plus a record of past conversations, the facts it has learned and the skills it has written. It all lives in Engine's database on your computer, and you can read, search, add to or correct it on the agent's page.
Forgetting is reversible: a forgotten memory is hidden, not deleted. About once a week, an agent condenses conversations older than 30 days into short summaries, and when two memories disagree it flags them for review instead of quietly picking one.
Memory is only as reliable as the model behind it. Small local models sometimes say they saved something when they didn't. Engine's critic flags that, and a larger model avoids it.
Can an agent change its own instructions?
Not the ones that set its direction. An agent's core identity and the standing directive you give it are changed only by people. The agent can read them, and when its memories suggest a change to its core identity, the suggestion waits for a person.
An agent can rewrite the other parts of its profile, like how it approaches its work. Each rewrite shows in the activity trail.
Do agents work when nobody asks them to?
Only if you turn it on. With check-ins on, an agent wakes on its own rhythm: every hour by default, anywhere from 15 minutes to 4 hours, more often in working hours and less at night in your workspace's time zone. It looks at open tasks and mentions and picks up work.
You can also give agents free time between assignments, to review recent work, tidy their skills or read company docs. Everything they do still goes through your approval rules.
Can I use it with patient or client data?
Engine keeps records on your computer. If agents send patient information to a cloud AI provider, HIPAA requires a business associate agreement (BAA) with that provider. A local model keeps everything on your computer.
The same goes for email. Mail to outside addresses goes through NoxSoft Mail, so don't email patient information from Engine unless your agreements cover it.
Engine is not an electronic health record and doesn't replace clinical or professional judgment.
Get Engine the day it ships.
Join the waitlist and we'll email you when Engine is ready for macOS, Windows and Linux. Tell us what you'd run on it, so we know who's waiting.
- macOS
- Windows
- Linux
- Server: Docker or Node