AgentML
AgentML is a universal rendering convention for AI agents. The same way browsers render HTML into a visual experience for humans, AgentML renders a complete agent-page/workspace for AI — context, state, permissions, actions, guidance, navigation — everything an agent needs to understand where it is and what it can do, in one response.
Quickstart
1. Install
pip install agentml
2. Setup in 3 Lines
from fastapi import FastAPI
from agentML import AgentML
app = FastAPI()
agent = AgentML(app)
agent.add_middleware() # Enables mutation response semantics
1. Mutation Enriched Response (POST /patients)
When an agent performs an action (e.g. POST /patients) and sends the header Accept: application/vnd.agentml+json, AgentML intercepts the response to tell the agent exactly what changed (delta), where the resource now lives (agent_resource), and what it can do next (actions) and where it can go (navigation):
POST /patients
Accept: application/vnd.agentml+json
Content-Type: application/json
{
"name": "Aanya Sharma",
"age": 34
}
Response:
{
"result": {
"id": 42,
"name": "Aanya Sharma",
"age": 34
},
"delta": {
"name": "Aanya Sharma",
"age": 34
},
"agent_resource": "/patients/42/agents",
"actions": [
{
"name": "UpdatePatientDetails",
"agent_endpoint": "/patients/42/agents"
},
{
"name": "ViewPatientDetails",
"agent_endpoint": "/patients/42/agents"
},
{
"name": "ArchivePatient",
"agent_endpoint": "/patients/42/agents"
}
],
"navigation": [
{
"label": "Appointments",
"agent_endpoint": "/patients/42/appointments/agents"
},
{
"label": "Billing",
"agent_endpoint": "/billing/agents"
},
{
"label": "Home",
"agent_endpoint": "/agents"
}
]
}
2. Resource Workspace (GET /patients/42/agents)
When an agent navigates to the resulting agent_resource, they receive a complete state-aware workspace:
GET /patients/42/agents
Response:
{
"agentML": "0.1",
"workspace": {
"title": "Patients Instance Agent Workspace",
"resource": "/patients/42",
"agent_resource": "/patients/42/agents",
"breadcrumb": ["Home", "Patients", "Patient 42"]
},
"identity": {
"role": "doctor",
"user_id": "doc_88"
},
"state": {
"id": 42,
"name": "Aanya Sharma",
"age": 34,
"archived": false
},
"capabilities": [
{
"name": "UpdatePatientDetails",
"description": "Modify name or age of an existing patient",
"fields": [
{
"name": "name",
"type": "string",
"required": true,
"current": "Aanya Sharma"
},
{
"name": "age",
"type": "integer",
"required": true,
"current": 34,
"constraints": {
"ge": 0,
"le": 120
}
}
]
}
],
"unavailable": [],
"navigation": [
{
"label": "Appointments",
"agent_endpoint": "/patients/42/appointments/agents"
},
{
"label": "Billing",
"agent_endpoint": "/billing/agents"
},
{
"label": "Home",
"agent_endpoint": "/agents"
}
],
"alerts": [],
"guidance": []
}
Understanding the Workspace Blocks
workspace: Native location context containing the human URL, the agent URL, and path breadcrumbs.identity: The dynamic security credentials and role context for the current request.state: The serialized current resource state fetched in-process.capabilities: Permitted business-language actions (never HTTP verbs) with dynamic schema validation rules and pre-filled current values.unavailable: Actions that are temporarily disabled due to the current resource state, complete with explanations.navigation: Context-aware sibling and sub-resource links matching the instance's active ID.alerts: Banner notifications or warning details generated dynamically based on state.
Philosophy: HTML for Humans, AgentML for AI
The web was built for human eyes. AI agents are currently forced to parse layout-heavy HTML, guess dynamic form flows, and reverse-engineer endpoints.
AgentML introduces a simple, universal web rendering convention: append /agents to any HTTP resource.
A browser requests a resource and gets human-friendly HTML. An AI agent requests the same resource + /agents (or issues mutations requesting the vnd.agentml+json payload) and gets a structured, action-ready Agent Workspace.
Core Features
- Relational Navigation: Context-aware nested resources (e.g.
/patients/42/agentslinks to/patients/42/appointments/agentssubstituting the identifier automatically). - Mutation Semantics (v0.4): Mutation writes return execution feedback, updated state deltas, and structured action and navigation lists to eliminate agent state blindness.
- Invisible-by-Design RBAC: Permissions are evaluated at render-time; capabilities the agent is unauthorized to perform are completely absent.
- Zero HTTP Verbs: All actions are presented in clean business terminology (e.g.
RegisterPatientinstead ofPOST /patients). - Fail-Closed Safety: Dynamic safety check hooks (
unavailable_fn) default to blocked state on exceptions. - Pydantic Validation: Auto-extracts field types, constraints (
ge,le, patterns, enums) and pre-fills current state values.
Metadata
Release files for agentml 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| agentml-0.1.0.tar.gz | 27.1 kB | Without provenance | Details for agentml-0.1.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| agentml-0.1.0-py3-none-any.whl | Python 3 | none | any | Details for agentml-0.1.0-py3-none-any.whl |
Total release size: 50.0 kB
Release files / agentml-0.1.0.tar.gz
| Download URL | agentml-0.1.0.tar.gz |
|---|---|
| Size | 27.1 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
17b2c039f527ba5eac10a446d796fdfc48266d7ec314cbcfb257bc33ed24e93c
|
|
BLAKE2b-256 checksum How to use checksums |
88b796fd56d8e07da709e64c67a075f9ca6287b165dc29ad555347e9d6df5d6f
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.6
|
Release files / agentml-0.1.0-py3-none-any.whl
| Download URL | agentml-0.1.0-py3-none-any.whl |
|---|---|
| Size | 22.9 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
f62a1ed1283d5f549e973d9e3916039ccb34d8bf32b85dfaa612481c66b62d97
|
|
BLAKE2b-256 checksum How to use checksums |
da5774bfb7819539c13d1d7b3b10754761e331b98ce81bd9d10320ca06a46b70
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.6
|
Security for agentml 0.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.