For forward-deployed engineers

Ship like the best FDE teams. Hand over what they never do: the ontology.

The forward-deployed model won: embed with the client, model their business, deliver working software fast. Its weak point is the handover — the ontology stays locked in a platform the client rents forever. ObjectStack makes the ontology typed, open files in the client’s repo; ObjectOS runs them governed, in our cloud or theirs.

ObjectOS product surface connecting business data, applications, and AI agents
AI Writes metadata Objects, permissions, workflows, tools
Human Reviews diff Business authority, data access, approvals
Runtime Enforces policy UI, APIs, audit, MCP, actions
16k tokens
A complete CRM — small enough for the client to review whole
Ontology-first
Model nouns and verbs before any screen is built
Client-owned
The definition lives in their repo, Apache-2.0

The deliverable

The handover is the product: an ontology in the client’s repo

A forward-deployed engagement ends. What remains decides what it was worth: typed objects, permissions, and flows that the client’s own team — and their AI agents — can read, run, and keep changing without you.

client-repo/
├── objectstack.config.ts   # the app manifest
├── src/objects/            # the client's nouns — typed, validated
├── src/flows/              # the client's verbs — approvals, automation
├── src/permissions/        # who may see and do what
└── dist/objectstack.json   # one artifact, runs on any ObjectStack runtime

The playbook

Ontology-first, like the model that proved it

The FDE method that works front-loads the business ontology before any AI application ships — generic models give generic answers; value lives in the client’s nouns and verbs.

Embed

1. Discover

Embed with the operators. Capture the objects, relationships, permissions, and approval paths that actually run the business.

Write

2. Model

Turn discovery into typed metadata with your coding agent — objects, flows, permissions as compact, validated definitions.

Run

3. Ship

The runtime derives the database, REST APIs, admin Console, and MCP tools from the definition. Demo on real workflows the same week.

Hand over

4. Govern and leave

Permissions, approvals, and audit are enforced by the runtime — the system stays inside its fence after you roll off.

Why open

The client keeps the ontology. You keep the client.

An open definition changes the economics of forward-deployed work: nothing about the engagement depends on reselling a platform seat.

No seat to sell

The framework is Apache-2.0. Your bid is your engineering, not a license markup the client resents later.

Security review reads the diff

Permissions and audit are metadata. The client’s security team reviews 2,000 readable lines, not a black-box platform boundary.

Any AI, under governance

The app is an MCP server out of the box — the client’s Claude, Cursor, or local models operate it inside user permissions.

Patterns compound

Every engagement leaves you reusable, typed patterns — objects, flows, permission sets — that your agent applies to the next client.

Decision surface

What changes, who reviews it, what runs

Engagement concernClosed platform (e.g. Foundry)ObjectStack / ObjectOS
Where the ontology livesInside the vendor’s platformTyped files in the client’s repo (Apache-2.0)
What the client pays forPlatform seats, indefinitelyOperations (ObjectOS) — or self-host at no license cost
Who maintains it laterVendor-trained specialistsAny coding agent plus a human reviewer
AI accessPlatform-native assistants onlyOpen MCP — Claude, Cursor, local models
Security reviewTrust the platform boundaryRead the diff — permissions and audit are metadata

Review checklist

FDE engagement checklist

  • Model the client’s nouns and verbs as objects and flows before any UI conversation.
  • Keep the whole definition context-sized, so your agent can reason about and refactor it whole.
  • Default permissions conservative; make every authority change explicit in the diff.
  • Hand over the repo, the compiled artifact, and a review checklist — not a login to your tenant.
  • Leave MCP enabled so the client’s own AI can operate the app under their permissions.

FAQ

Questions this page should answer

Is this a Palantir Foundry replacement?

For the ontology-first application workflow — modeling a business and shipping governed apps on it — the overlap is real, and the definition layer here is open instead of proprietary. Foundry remains stronger for massive-scale data federation and analytics pipelines; many clients need only the application layer.

Do I need ObjectOS to serve clients?

No. The open-source ObjectStack runtime self-hosts with no license, Console and governance included. ObjectOS is the commercial platform for clients who want it operated — the browser-based AI Builder, cloud or Enterprise deployment, and support.

Next pages

Keep building the evaluation packet.