SmallForce Enterprise

The AI workforce platform that runs in your cloud.

AI employees for every department, an internal app platform that hosts a million apps on one server, and search across everything your company knows. Deployed in your account, signed in with your identity provider.

northwind-prod · your-cloud · us-east-1 Example deployment
AI employees
1429 departments
Internal apps
1,28437 running now
Indexed
18.2Mpassages
Approvals
6waiting on people

Services

  • smallforce-os AI employees and workspace 3 nodes Healthy
  • app-platform Internal apps, isolated per app 2 nodes Healthy
  • search-index Hybrid keyword + vector index 2 nodes Indexing 64%
  • identity Okta · OIDC — Connected

Live activity

  • Maya resolved 18 support tickets2m
  • Theo deployed “Vendor onboarding”6m
  • Daniel queued 14 invoices for approval11m
  • Nora drafted the Contoso renewal19m

Needs approval · Daniel

Pay 14 supplier invoices, $182,400

1M+
apps deployed on a single server
< 100 ms
search over a million passages
Yours
the cloud account every file is stored in
OIDC
sign-in with your identity provider

Northwind Group is a sample company. Search figure from our benchmark on one index. Model calls go to the AI providers you choose.

App platform

Your own Lovable, for the whole company.

Anyone describes the tool they need. An AI employee builds it, ships it behind your company login and keeps it running.

1M+ apps deployed on a single server.

Apps load the moment someone opens them and leave memory when they go quiet. You pay for the apps in use, not every app that exists.

Running nowDeployed, idleEach square is 1,000 apps. Illustrative.

Internal apps1,284 apps · 37 running

AppTeamAccessTrafficState
vendor-onboardingProcurementOkta SSO1.2k/min Running
field-quotesSalesOkta SSO310/min Running
stock-checkOperationsTeam only— Idle
holiday-rotaPeopleGoogle list— Idle
warranty-claimsSupportOkta SSO88/min Running

Theo · Web developer“Vendor onboarding” is on Preview with Okta sign-in, a database for submissions and storage for tax forms. Promote to Production?

Release log · vendor-onboarding

  1. 12:04:11buildrelease r14 · 18 modules · sha256 verified
  2. 12:04:12previewr14 live at vendor-onboarding--preview
  3. 12:04:12accesspreview limited to org members
  4. 12:06:40promoter14 → production, no rebuild
  5. 12:06:40routeproduction serves r14 · r13 kept for rollback
Isolation
Every app in its own V8 isolate
Data
A SQLite database per environment
Storage
Files and objects per app
Background
Queues, workflows, scheduled jobs
Releases
Immutable, Preview first, instant rollback
Access
OIDC SSO, team only, Google list, password
Secrets
AES-256-GCM, never in the app bundle
Logs
Requests, errors, 365 days

Full-stack apps

Not prototypes. Real software, with a real backend.

Every app your team asks for ships with its own database, file storage, background jobs and access to your company's tools. Built by an AI employee, owned by you.

src/server.ts wrangler.jsonc written by Theo
export default {  async fetch(request, env, ctx) {    // Signed in with your identity provider    if (!ctx.user) return new Response("Sign in", { status: 401 });    const form = await request.formData();​    // Your own SQL database, one per environment    const { rows: [vendor] } = await env.DB.query(      "INSERT INTO vendors (name, owner) VALUES (?, ?) RETURNING id",      [form.get("name"), ctx.user.email],    );​    // File storage with signed uploads    const { upload } = await env.STORAGE.createUpload({      path: `vendors/tax-forms/${vendor.id}.pdf`,      contentType: "application/pdf",      sizeBytes: Number(form.get("size")),    });​    // Any connected app, without ever seeing its credentials    await env.INTEGRATIONS.execute({      connectionId: env.GMAIL_CONNECTION,      toolKey: "GMAIL_SEND_EMAIL",      arguments: { to: "ap@northwind.com", subject: "New vendor" },    });​    // Background work: a queue and a durable workflow    await env.VENDOR_CHECKS.send({ vendorId: vendor.id });    await env.ONBOARDING.create({ params: { vendorId: vendor.id } });​    return Response.json({ upload });  },​  // Runs every night at 2am  async scheduled(controller, env) {    const { defaultModelId } = await env.AI.listModels();    await env.AI.createResponse({      model: defaultModelId,      input: "Summarize today's vendor sign-ups for procurement.",    });  },};
Server code for the Vendor onboarding app, using the database, file storage, integrations, a queue, a workflow and AI.

What Theo shipped

One app. Seven platform services. Zero servers.

A signed-in vendor form with its own database, file storage, a background check queue, an onboarding workflow, a nightly AI summary and a Gmail connection.

Vendor onboardingproduction · release r14 Live
  • Databaseenv.DB · vendors · tax_forms · checks 12.4 MiB of 100 MiB
  • File storageenv.STORAGE · 1,204 objects 186 MiB of 500 MiB
  • Queuevendor-checks 3 in flight · 1,920 today
  • Workflowonboarding 12 running · 408 completed
  • Schedule0 2 * * * Next run tonight at 02:00
  • IntegrationsGmail · Salesforce 318 calls today
  • AIenv.AI 64 responses today
  • Secrets4 secrets · 2 variables Production and Preview

Sample usage. Standard Worker APIs your team can read, change or take with them.

Built in

Everything a production app needs. Nothing to provision.

The same backend primitives your engineers reach for, available to every app from its first release. No database to set up, no queue to run, no keys to paste.

Data

  • SQL database

    env.DB.query(sql, params)

    A SQLite database for every environment. Preview data never touches Production.

    100 MiB default · 10k rows per result

  • File storage

    env.STORAGE.createUpload(file)

    Signed uploads and downloads that go straight from the browser to storage.

    500 MiB · 10k objects · 25 MiB a file

  • Static assets

    env.ASSETS

    Pages, scripts and images served from the immutable release.

    Versioned with every release

Background work

  • Queues

    env.VENDOR_CHECKS.send(msg)

    Batch consumers with retries for slow or bursty work.

    Per app and environment

  • Durable workflows

    env.ONBOARDING.create(run)

    Multi-step processes that survive restarts and resume where they stopped.

    Per app and environment

  • Scheduled jobs

    scheduled(controller, env)

    Cron triggers for nightly reports, syncs and clean-ups.

    triggers.crons

AI and connections

  • AI models

    env.AI.createResponse(req)

    70+ curated models called server-side, with no API key in your code.

    Or your private models

  • Your company's apps

    env.INTEGRATIONS.execute(call)

    Gmail, Salesforce, Stripe and 200+ more, plus your OpenAPI specs and MCP servers.

    Credentials never reach app code

  • Any other API

    fetch(url)

    Call internal or external services, limited to the destinations you allow.

    Outbound allowlist per app

Identity and operations

  • Signed-in user

    ctx.user

    The verified person from your identity provider, on every request.

    id · email · name · method

  • Secrets and variables

    env.SALESFORCE_TOKEN

    Write-only secrets per environment, injected at runtime and never bundled.

    AES-256-GCM

  • Logs and events

    env.TELEMETRY

    Structured events and request logs, searchable from the dashboard.

    Kept 365 days

Run them properly

Operated like production software, from day one.

Every app comes with the dashboards and controls your platform team would otherwise build: releases, logs, traffic, and security settings per app.

Sample app and figures. The tabs and settings are the product's own.

  • Preview, then Production

    Every change lands on Preview. Promote the exact release you checked, with no rebuild.

  • Rollback in one click

    Every release is immutable. The previous one is always a click away.

  • Rate limits per app

    Protect internal systems from a runaway script or a noisy team.

  • Your own environments

    Add Test or Staging next to Preview and Production, each with its own data.

Private AI models

Your own models, trained on your business, running in your cloud.

Use the 70+ models we host, or deploy open-weight models privately and fine-tune small ones on your own work.

Modelsnorthwind-prod

ModelBaseRuns onStateUsed by
northwind-support-8b Fine-tuned · Llama 3.1 8B Your cloud · Baseten Serving
northwind-contracts-32b Fine-tuned · Qwen3 32B Dedicated · Fireworks Serving
gpt-oss-120b Open weights Your cloud · Baseten Serving
deepseek-v3.1 Open weights Dedicated · Fireworks Standby —

Fine-tuning northwind-support-8b · v4step 2,400 / 3,600

18,240 resolved ticketsLlama 3.1 8B baseYour cloud · Baseten

Sample deployment. Northwind models and figures are illustrative.

  • Private inference

    Open models run inside your own cloud, so prompts, documents and answers never leave it.

  • Trained on your work

    Fine-tune on your tickets, contracts and calls. The model learns your products, policies and tone.

  • Small models, smaller bills

    A tuned small model handles the routine work. Frontier models step in only when a task needs them.

Company brain Search is the start. Meet the company brain. It learns from your team's work, keeps its own notes and answers with sources. How it learns

AI employees

A workforce for every department, governed like one.

Hundreds of AI employees across your teams, each with its own role, model and permissions, and one place to see what all of them are doing.

Workforcethis week

DepartmentEmployeesTasks doneWaitingDefault model
Customer support 41 3,912 2 Claude Sonnet 5
Sales 28 1,204 1 GPT-5.5
Finance 17 846 3 Claude Opus 5
Operations 22 1,530 0 Gemini 3.8 Flash
Marketing 19 702 0 GPT-5.5
IT 15 388 0 northwind-support-8b

Policy

employee
daniel · finance
folders.finance/
read-write
folders.hr/
no-access
internet
allowed-domains
allow
stripe.com, netsuite.com
commands
require-approval
approval.over
$50,000

Set in the dashboard. Enforced by the OS.

  • Sandboxed by the operating system

    Folder, network and command rules are enforced below the model, so a clever prompt can't talk its way past them.

  • Approvals above your line

    Anything over the threshold you set waits for a person. Everything else gets done.

  • A model per role

    70+ models from 24 labs, chosen per employee, or your team's own ChatGPT and Grok plans.

Sample company. Departments and figures are illustrative.

Security

Built for the security review, not around it.

Controls live in the product, where your team can see and test them. The questionnaire, architecture notes and DPA are ready when you are.

Identity

Your login, everywhere

Internal apps sign people in through your identity provider. Owners, admins and members control the rest.

Company SSOConnected

Provider
Okta
Protocol
OpenID Connect
Apps protected
23
Last change
Priya K. · 3 days ago

Isolation

Nothing shared by accident

Every app and every AI employee runs in its own sandbox, with its own data and only the access you grant.

vendor-onboarding

Code
Own V8 isolate
Database
Own SQLite cell per environment
Files
Own storage prefix, signed URLs
Credentials
Held by the platform

Secrets

Write-only, encrypted

Secrets are encrypted with AES-256-GCM, injected at runtime and never shown again, not even to admins.

Secretskeyring v2

  • SALESFORCE_TOKENproduction••••••••
  • SUPABASE_SECRET_KEYproduction••••••••
  • SUPABASE_SECRET_KEYpreview••••••••
Audit trail
Employee actions, integration calls, releases, SSO changes
Approvals
Thresholds per employee and action
Retention
App logs and requests for 365 days
Training
Your data never trains anyone's models
Network
IP allowlists and outbound allowlists per app
Request the security pack Questionnaire · architecture notes · DPA

Deployment

Your cloud, or ours. The same platform either way.

Every component ships as containers you can run in your own account. Or we run a dedicated deployment for you.

What runs in your account

  • smallforce-osAI employees, workspace, schedules
  • app-platformHosts and isolates your internal apps
  • app-controlReleases, environments, access policies
  • search-indexHybrid index over your documents
  • identityOIDC connection to your provider
SmallForce Cloud compared with running SmallForce in your own cloud
SmallForce CloudYour cloud
Where it runsA dedicated machine we operateYour AWS, GCP or Azure account
Your dataIsolated to your companyStored only in your account
UpgradesAutomaticScheduled with your team
NetworkOur managed networkYour VPC and firewall rules

Model calls go to the AI providers you choose.

  1. Pilot

    Start with one department

    Real work on a real team, with success agreed up front.

  2. Roll out

    Connect the company

    SSO, permissions and data sources, team by team.

  3. Operate

    Run it with us

    A shared channel and regular reviews of what the team is doing.

Talk to sales

Put an AI workforce inside your company.

Tell us what you want to run. We'll come back with a pilot plan, an architecture review and pricing for your deployment.

Pricing
Per deployment, quoted after a call
Pilot
One department to start
Security pack
On request

What do you want to run?

Or book a call.