Authentication, routing, and limits
The gateway accepts a request, resolves the caller and organization context, routes it to the owning service, and applies configured limits.
Install the ERP.AI skills by running this terminal command: curl -fsSL https://apps.erp.ai/api/install | bash When it finishes, reload your skills so they're available, then read https://apps.erp.ai/api/skills.json to see what each one does. Ask which business system I want to use and what work I need done. Help me connect an existing workspace or choose a system. Read the installed skills to discover the actions available in that system, then help me operate it: read or update records, prepare reports, or run supported workflows. Respect the workspace's permissions and approval settings. If I need a new system, plan it with me first, then use the build-app skill to build it. If we need an API key, walk me through creating one at https://apps.erp.ai/apps?settings=api-keys — the installed skills explain where to store it.
One business core.
Every way of working.
One business core for your team, your agents, and the apps you build. Shared data and business logic, through APIs. Each with its own access.
Headless SaaS separates business logic from the interface. The interface is the “head”. Data, workflows, and rules live in a shared backend, accessible through APIs.
Add a customer portal, connect an integration, or bring in an AI agent. Each uses the same business capabilities. Change the experience without rebuilding the core.
Your team, an agent, or an integration. The same business, with access that belongs to each.
Build apps, work with records, run workflows, and read reports in a complete web product. Your team can start here and keep working here as you connect more ways into the platform.
Every client connects through the gateway. The service that owns the operation checks access and applies the business rules. Change the interface. Keep the foundation.
Work with records, workflows, and reports in ERP•AI. The gateway identifies your team; the owning services check access and carry out each operation.
The gateway accepts a request, resolves the caller and organization context, routes it to the owning service, and applies configured limits.
The App Builder API owns apps, schemas, records, views, forms, workflows, reports, documents, pages, hosting metadata, and audit events. PostgreSQL is the durable source of truth.
ERP•AI’s identity service handles organizations, users, service accounts, keys, OIDC, SCIM, hosted-app sign-in, domains, and portals. Authentication and enforcement stay with the services; the Permission API owns roles, permissions, assignments, and invitations.
The Agent API handles conversations, knowledge, app plans, reports, documents, pages, search, and action approvals. The Renderer serves multi-tenant hosted pages and portals.
Start with the web app. Everything you build becomes part of the same app model, ready for the clients you connect next.
Explore ERP•AICreate apps with typed tables, records, references, formulas, views, forms, layouts, and collaborative updates.
Build workflows, inspect execution state and node output, see errors, stop runs, and retry supported runs.
Create reports, documents, templates, dashboards, and custom pages on the same app model.
Manage organizations, users, roles, permissions, invitations, service accounts, keys, OIDC, and SCIM.
Serve hosted pages, portals, domains, and external apps through the platform's identity and hosting services.
Use schema branches, commits and merges, snapshots and restore, audit history, workflow run history, trash recovery, and saved page versions where supported.
Keep your business rules consistent as new clients connect. Whether a request comes from a person or software, the platform identifies the caller and sends the work to the service responsible for it.
The gateway identifies the caller and organization from a sign-in or service credential. Every request starts with that context.
The service responsible for the resource checks the caller’s access to the requested app, record, workflow, page, or other resource.
The owning service validates the request and reads or writes the authoritative data. The rules stay with the operation, wherever it starts.
Supported product areas keep audit events, versions, workflow runs, snapshots, or deployment history, giving you a way to inspect the changes they own.
Tools to act. Context to reason. Somewhere to run. A record to learn from. Headless APIs connect an agent to your business; these related ERP•AI projects support the work around it.
Brief Proto. It works across your files and tools, then brings the results back. It runs on ERP•AI’s managed plan by default, or on your own key from a provider such as OpenAI, Anthropic, Google Gemini, OpenRouter, or Fireworks.
Proto works with your files and the web on its own. With ERP•AI connected, it can also build and update apps, records, workflows, and reports through the same headless APIs as your web app.
Who is asking, what are they trying to do, and how much access does that task need? Herbert brings a small, local model to those questions. Built from the local decision-model code behind SmallDecide, the prototype reviews access requests and recommends scope and review routes for people, agents, service accounts, and hosted apps.
The first experiment is an agent passport: propose read-only access for a short task instead of carrying broad permissions into every job. Herbert recommends; the identity service verifies the caller, approves the scope, and enforces access. Temporary passports, permission previews, and task-bound access are the next directions to explore.
SmallDecide runs locally to review a proposed action, route a request, or score against a rubric. It returns structured choices, scores, and probabilities rather than a chat response.
A narrow decision does not always need a large generative model. Your application supplies the context and rules, then decides how to use the result alongside its permissions and approvals.
An agent needs an isolated workspace for code, files, and dependencies. But it may spend much of its time waiting for a person or an API. Keeping a machine running for every waiting agent ties up resources; starting over loses its working state.
vmyard is being built to schedule sandboxes across hosts and manage their run, sleep, and wake cycle. Placement accounts for snapshots and wake-up costs, with the aim of letting agents resume work while sharing capacity efficiently.
FFS brings records, relationships, vector search, and durable writes into one database engine. An agent can retrieve similar material, follow its connections, and write back a result.
Related facts and embeddings need to stay in step. A shared transaction boundary helps agent applications keep the result and the context behind it together. FFS is a related context database; PostgreSQL remains ERP•AI’s authoritative business-record store.
nogrep follows calls, imports, and other code relationships to retrieve a relevant slice of a repository. Its tools let an agent ask what calls a function or what a change could affect.
Coding agents have a limited context budget. Structural retrieval helps spend it on connected code instead of repeatedly searching and reading whole files.
TraceBreak links plans, tool calls, observations, and memory into an inspectable event history. Replay can compare model decisions under a different model or prompt and reveal the first changed step.
When an agent goes wrong, the final answer rarely explains why. A connected trace lets a team follow the evidence and compare behavior; replay does not execute tool calls by default.
The practical detail behind the shared core.
Tables, columns, records, views, forms, workflows, reports, documents, and pages remain part of the same app.
People and configured software clients use identities the platform can check and attribute.
Organization, app, role, resource, and action checks stay with the services that own them.
PostgreSQL is the authoritative store; an accelerated record path falls back to PostgreSQL when needed.
Validation and app behavior live in the platform rather than being reimplemented in each client.
Supported workflows keep execution state, output, errors, stop and retry controls, and summaries.
Branches, commits, merges, snapshots, audit events, page versions, and deployment history cover the changes they own.
Pages, portals, external apps, domains, and their identity settings can be managed as product surfaces.
The web app, Proto, hosted output, and integrations can evolve without moving business ownership out of the platform services.
The web app is for people who want to work directly with tables, forms, reports, workflows, documents, and settings. Programmatic clients are useful for repeatable work, integrations, hosted experiences, and agent-assisted tasks.
Both paths still need identity, tenant context, resource permissions, validation, and an owning service.
Connect software with its own credentials and assigned permissions. Your team, an agent, and a portal can use the same platform with different levels of access.
Choose the operations your client needs, then use the platform documentation to configure its credentials and requests. Custom clients keep their own interface and release cycle.
Turn repeatable work into workflows you can inspect. Follow run state and output, see where an error occurred, and stop or retry supported runs.
The interface changes; the owning services and permission checks do not.
Each caller is identified and checked instead of borrowing a person's browser session.
Business rules stay in the service instead of being copied into every client.
People can build and run apps through tables, forms, dashboards, reports, workflows, documents, custom pages, and administration screens. Programmatic access does not replace that product.
The gateway resolves identity and tenant context. The service that owns the requested resource checks access before it performs the operation.
Proto, hosted pages, external apps, and approved integrations can use the platform services without recreating ERP•AI business logic in each client.
ERP•AI gives you a complete web product and a service layer that other clients can use. Your data model and business logic can serve a team workspace, a portal, an agent, or an integration.
Clients connect through supported operations with their own assigned access. That keeps the core reusable while allowing each experience to evolve on its own.
What headless changes.
And what it keeps.
Headless SaaS makes business capabilities available independently of a single interface. Traditional SaaS can offer APIs too; the useful distinction is how much work another client can do through them. ERP•AI combines a complete web app with shared services that other clients can use through supported operations.
No. ERP•AI includes a complete web app for tables, forms, reports, workflows, documents, and administration. Headless means you can also work through another interface, an agent, or an integration. You can use the included product from day one.
Yes. Proto is a personal agent for your business that works across your files and tools. Its ERP•AI actions work with apps and their resources. Integrations connect through supported APIs with their own credentials. Each caller’s permissions determine the work it can do.
No. A team member, an agent, and a customer portal can have different permissions. The gateway identifies the caller and organization, then the owning service checks access to the requested resource. Headless access exposes supported operations, not every internal function.
When your business needs more than one way to get work done: a workspace for your team, a portal for customers, an agent for supported tasks, or an integration with another system. Reusing the core keeps you from rebuilding the same logic for each. Start with the web app when it fits; add custom clients when their value justifies the integration and maintenance work.
Start with the ERP•AI web app. Bring in Proto, connect your tools, or build an interface of your own. Keep the business you build ready for the way you want to work next.
Retrieved through ERP•AI with the agent’s assigned access.
Welcome back,
Acme Studio.
A focused view of your business, with its own sign-in and access.
{
"customer": "Acme Studio",
"owner": "Alex",
"status": "active"
}
{
"customer": "Acme Studio",
"owner": "Alex",
"status": "active"
}
We use essential cookies to run the site and optional cookies for features, analytics, and relevant content. See Cookie policy
We use cookies to enhance your experience, analyze site traffic, and serve relevant content. By clicking "Accept All," you agree to our use of cookies. You can customize your preferences at any time.
Your browser's Global Privacy Control signal is honored: advertising cookies are off.
Learn more about how we use cookiesThese cookies are required for the website to function properly. They ensure security, enable basic features like page navigation, and store user session data. You cannot disable these cookies.
These cookies enable additional features that enhance your experience, such as live chat, video playback, personalized content recommendations, and remembering user preferences.
These cookies help us understand how visitors interact with our site by collecting anonymous usage data. This allows us to measure performance, detect issues, and continuously improve the user experience.
These cookies allow us and advertising partners, including X, to deliver ads tailored to your interests. They track browsing habits across sites to provide relevant advertising and measure ad effectiveness.