Skip to main content

Secure API gateway for AI-generated artifact apps — external proxy, internal API forwarding, and user-isolated database access with JWT-scoped auth tokens.

Project description

artifact-gateway

PyPI version Python versions License CI

Secure API gateway for AI-generated artifact apps — external HTTPS proxy, internal API forwarding, and user-isolated database access with JWT-scoped auth tokens.

artifact-gateway demo — an AI-generated interactive dashboard pulling live data and rendering it in the browser, routed through the gateway

Above: an AI-generated artifact (a live FIFA forecast dashboard) calls external APIs and isolated data through artifact-gateway — no secrets in the page, no raw network access.


What it is

AI agents (Claude Code, Codex, Gemini CLI) can produce interactive web apps — artifacts — that need to call external APIs, read internal data, and persist state. These apps run inside iframes and cannot authenticate directly.

artifact-gateway provides:

  • App tokens — short-lived JWTs scoped to a user+context, derived from the user's RBAC role.
  • External proxy — forward calls from artifact apps to any external HTTPS API without exposing the user's credentials.
  • Internal proxy — forward calls to internal OhWise APIs with allowlist enforcement; the user's JWT is forwarded so RBAC is handled by the target service.
  • DuckDB handler — per-user and per-session isolated DuckDB files on the local filesystem.
  • MongoDB handler — per-user and per-session isolated MongoDB collections via Motor.

Browser client SDK: the matching client that artifact apps load in the iframe lives in sdk/js/ and is published to npm as artifact-sdk (npm i artifact-sdk, or via CDN). This Python package is the server half; the JS SDK is the client half.


Installation

Base install (token + proxy only, no database drivers):

pip install artifact-gateway

With DuckDB support:

pip install "artifact-gateway[duckdb]"

With MongoDB (Motor) support:

pip install "artifact-gateway[mongo]"

Full install:

pip install "artifact-gateway[all]"

Quick start

1. Issue an app token

from artifact_gateway import issue_app_token, scope_from_role

token = issue_app_token(
    secret_key="your-service-secret-key",
    user_id="user-abc123",
    account_id="account-xyz789",
    context="lab_session:sess-001",
    scope=scope_from_role("member"),  # derives scopes from RBAC role
    ttl=14400,  # 4 hours (default)
)

2. Validate a token on incoming requests

from artifact_gateway import validate_app_token

try:
    claims = validate_app_token(secret_key="your-service-secret-key", token=token)
    user_id = claims["user_id"]
    scope = claims["scope"]
except ValueError as exc:
    # expired or tampered token
    raise

3. Proxy an external API call

from artifact_gateway import ExternalProxy

proxy = ExternalProxy()

result = await proxy.call(
    scope=claims["scope"],
    method="GET",
    url="https://api.example.com/v1/data",
    headers={"X-Api-Key": "..."},
)
# result = {"status": 200, "headers": {...}, "body": {...}}

4. Proxy an internal OhWise API call

from artifact_gateway import InternalProxy

proxy = InternalProxy(base_url="http://localhost:8000")

result = await proxy.call(
    scope=claims["scope"],
    user_jwt="<user's original OhWise JWT>",
    path="/api/agent",
    method="GET",
)
# result = {"status": 200, "body": {...}}

5. DuckDB — user-isolated database

from artifact_gateway.db.duckdb import DuckDBHandler

handler = DuckDBHandler(workspace="/var/ohwise/workspaces")

# Create and populate a table.
await handler.exec("user", user_id, "analytics", "CREATE TABLE events (ts TEXT, val INT)")
await handler.exec("user", user_id, "analytics", "INSERT INTO events VALUES (?, ?)", params=["2026-01-01", 42])

# Query it.
result = await handler.query("user", user_id, "analytics", "SELECT * FROM events")
# {"columns": ["ts", "val"], "rows": [["2026-01-01", 42]], "rowcount": 1}

Session-scoped database (isolated to one Lab session):

await handler.exec("session", user_id, "scratch", "CREATE TABLE t (n INT)", session_id="sess-001")

6. MongoDB — user-isolated collections

import motor.motor_asyncio
from artifact_gateway.db.mongo import MongoHandler

client = motor.motor_asyncio.AsyncIOMotorClient("mongodb://localhost:27017")
db = client["ohwise"]
handler = MongoHandler(db=db)

# Find documents.
docs = await handler.find("user", user_id, "notes", filter={"status": "active"})

# Upsert.
await handler.upsert("user", user_id, "notes",
    filter={"key": "my-note"},
    update={"$set": {"content": "updated content"}},
)

# Delete.
await handler.delete("user", user_id, "notes", filter={"archived": True})

7. Credential vault — reference secrets by name

Artifact apps should never embed plaintext API keys. Store the secret encrypted, then have the app reference it by name; the gateway resolves and injects it on the outbound request.

from artifact_gateway import CredentialVault, apply_credential, PLACEMENT_BEARER

vault = CredentialVault(master_key="your-encryption-key")

# At save time (store ciphertext in your DB):
ciphertext = vault.encrypt("sk-live-abc123")

# At request time, resolve and inject:
secret = vault.decrypt(ciphertext)
headers, query = apply_credential(secret=secret, placement=PLACEMENT_BEARER)
# headers = {"Authorization": "Bearer sk-live-abc123"}

placement can be "bearer", "header" (with name="X-Api-Key"), or "query" (with name="api_key").

8. Files — user/session-isolated read & write

from artifact_gateway import FilesHandler

files = FilesHandler(root="/var/ohwise/workspaces")

await files.write("user", user_id, "exports/report.csv", "a,b\n1,2\n")
out = await files.read("user", user_id, "exports/report.csv")
await files.list("user", user_id, "exports")
await files.delete("user", user_id, "exports/report.csv")

Paths are confined to the user's (or session's) directory; traversal outside it raises ValueError.

9. Streaming external proxy

proxy = ExternalProxy()
async for chunk in proxy.stream(
    scope=claims["scope"], method="POST",
    url="https://api.anthropic.com/v1/messages",
    headers={"x-api-key": "..."}, body={...},
):
    relay_to_client(chunk)  # e.g. as SSE

10. Refresh a token

from artifact_gateway import refresh_app_token

fresh = refresh_app_token(secret_key, old_token)  # same claims, new exp

Concepts

App token

A short-lived JWT (HS256) signed with the service's SECRET_KEY. Claims include user_id, account_id, context (e.g. lab_session:<id>), and a scope list derived from the user's RBAC role.

The token is injected into artifact app iframes via postMessage by the parent frame. It is stored in memory only — never in localStorage or cookies.

Scope claims

Scope Meaning
external:* Call any external HTTPS API
internal:read Read from internal OhWise APIs (GET)
internal:write Mutate internal OhWise APIs (POST/PUT/DELETE)
db:user:read Read from user-scoped databases
db:user:rw Read and write user-scoped databases
db:session:read Read from session-scoped databases
db:session:rw Read and write session-scoped databases
db:shared:read Read from shared databases (admin+ only)

Role → scope mapping

Role External Internal write DB user rw DB shared read
viewer No No No No
member Yes Yes Yes No
admin Yes Yes Yes Yes
system_admin Yes Yes Yes Yes
platform_owner Yes Yes Yes Yes

Data isolation

  • DuckDB: each user's files live under {workspace}/{user_id}/db/. Session-scoped files live under {workspace}/{user_id}/sessions/{session_id}/db/. The proxy never allows path traversal beyond the user's directory.
  • MongoDB: collection names are prefixed automatically — app_user_{user_id}_{collection} or app_session_{session_id}_{collection}. User code never sees the real collection name.

Integrating with FastAPI

Wire the proxy handlers into a FastAPI router alongside token validation middleware:

from fastapi import APIRouter, Depends, HTTPException, Header
from artifact_gateway import (
    validate_app_token,
    ExternalProxy,
    InternalProxy,
    ExternalCallRequest,
    InternalCallRequest,
)

SECRET_KEY = "your-secret-key"
router = APIRouter(prefix="/api/lab/app")
external_proxy = ExternalProxy()
internal_proxy = InternalProxy(base_url="http://ohwise-backend:8000")


async def get_claims(authorization: str = Header(...)) -> dict:
    token = authorization.removeprefix("Bearer ")
    try:
        return validate_app_token(SECRET_KEY, token)
    except ValueError as exc:
        raise HTTPException(status_code=401, detail=str(exc))


@router.post("/external")
async def proxy_external(req: ExternalCallRequest, claims: dict = Depends(get_claims)):
    try:
        return await external_proxy.call(
            scope=claims["scope"],
            method=req.method,
            url=req.url,
            headers=req.headers,
            body=req.body,
        )
    except PermissionError as exc:
        raise HTTPException(status_code=403, detail=str(exc))


@router.post("/internal")
async def proxy_internal(req: InternalCallRequest, claims: dict = Depends(get_claims)):
    try:
        return await internal_proxy.call(
            scope=claims["scope"],
            user_jwt=claims["_user_jwt"],  # store at token issuance
            path=req.path,
            method=req.method,
            body=req.body,
            query=req.query,
        )
    except PermissionError as exc:
        raise HTTPException(status_code=403, detail=str(exc))

Contributing

  1. Fork the repository.
  2. Create a feature branch.
  3. Run tests: pip install -e ".[dev]" && pytest.
  4. Submit a pull request.

License

Apache 2.0 — see LICENSE.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

artifact_gateway-0.3.1.tar.gz (3.8 MB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

artifact_gateway-0.3.1-py3-none-any.whl (30.6 kB view details)

Uploaded Python 3

File details

Details for the file artifact_gateway-0.3.1.tar.gz.

File metadata

  • Download URL: artifact_gateway-0.3.1.tar.gz
  • Upload date:
  • Size: 3.8 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.0

File hashes

Hashes for artifact_gateway-0.3.1.tar.gz
Algorithm Hash digest
SHA256 dc06b6c0e383c875c906dd4aec9cd43801d1fabe4f4a4a0cfae35049062c3e0c
MD5 88be5a137553bdd3b435310d8d63d4de
BLAKE2b-256 1981e08460d456df5cc52bf1cae7889a8ffb5cf21da67e8ffc897b3f4ad4c004

See more details on using hashes here.

File details

Details for the file artifact_gateway-0.3.1-py3-none-any.whl.

File metadata

File hashes

Hashes for artifact_gateway-0.3.1-py3-none-any.whl
Algorithm Hash digest
SHA256 45e7ef5060589a8986993db6f89133032bccad5f15258f4bdeb2a75dfd1697e6
MD5 2b372dc9fe9acf0709a377375936c464
BLAKE2b-256 a2d0d8d0aa400a341ec392b6383f40ebe89573727eccb9d9bd02a6cc479388f8

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page