algomancy-api

FastAPI HTTP interface that exposes the same scenario- and data-management surface used by algomancy-gui, so a remote frontend (browser SPA, native desktop app, another Python process) can drive an Algomancy backend over the network instead of importing it in-process.

The HTTP layer is deliberately thin: every route maps to a single ScenarioManager / SessionManager method and responses use the existing to_dict() payloads. There is no parallel domain model — clients work with the same Scenario, DataSource, and KPI concepts the Dash GUI does.

Quick start

from algomancy_api import ApiConfiguration, ApiLauncher
from algomancy_data import DataSource

cfg = ApiConfiguration(
    etl_factory=MyETLFactory,
    kpis=kpis,
    algorithms=algorithms,
    schemas=schemas,
    data_object_type=DataSource,
    has_persistent_state=True,
    data_path="data",
    autocreate=False,
    autorun=False,
)
app = ApiLauncher.build(cfg)  # returns a standard FastAPI app
ApiLauncher.run(app)          # blocks; uses cfg.host / cfg.port

The server starts (default 127.0.0.1:8051) and serves the OpenAPI schema at /openapi.json plus the interactive Swagger UI at /docs. All scenario/data endpoints live under /api/v1/sessions/{session_id}/....

ApiLauncher.build returns a standard FastAPI instance — for production deploys you can hand it to your own uvicorn / gunicorn process manager instead of using ApiLauncher.run.

To try it against the bundled example wiring:

from algomancy_api import ApiLauncher
from algomancy_api.example import build_example_config

ApiLauncher.run(ApiLauncher.build(build_example_config()))

Endpoint inventory

All routes are prefixed with cfg.prefix (default /api/v1).

Sessions

• GET /sessions — list [{id, display_name}, ...] and the default id
• POST /sessions — create a new session — body {"display_name": "..."}
• POST /sessions/{sid}/copy — copy — body {"new_display_name": "..."}
• PATCH /sessions/{sid} — rename — body {"display_name": "..."}
• DELETE /sessions/{sid} — delete a session and all its data; the last remaining session is auto-replaced with a fresh "main"

{sid} accepts either the session's UUID (canonical) or its current display_name (soft-compat alias).

Algorithm + KPI discovery

• GET /sessions/{sid}/algorithms — list algorithm names
• GET /sessions/{sid}/algorithms/{name}/parameters — per-parameter descriptors
• GET /sessions/{sid}/kpis — list KPI template names

Scenarios

• GET /sessions/{sid}/scenarios — list scenarios
• POST /sessions/{sid}/scenarios — create a scenario
• GET /sessions/{sid}/scenarios/{id} — full scenario incl. KPIs and result
• DELETE /sessions/{sid}/scenarios/{id} — delete a scenario
• POST /sessions/{sid}/scenarios/{id}/run — enqueue for processing
• GET /sessions/{sid}/scenarios/{id}/status — lightweight status/progress (polling)
• GET /sessions/{sid}/processing — currently running scenario, or null

Data

• GET /sessions/{sid}/data — list dataset keys
• GET /sessions/{sid}/data/{key} — parsed JSON of a dataset
• DELETE /sessions/{sid}/data/{key} — delete a dataset
• POST /sessions/{sid}/data/{key}/derive — derive a new dataset
• POST /sessions/{sid}/data/from-json — add a dataset from DataSource.to_json() payload
• POST /sessions/{sid}/etl — run ETL over a multipart upload

Meta

• GET /health — liveness probe
• GET /openapi.json, GET /docs — OpenAPI schema and Swagger UI

Polling pattern

import time
import httpx

base = "http://127.0.0.1:8051/api/v1"
session = httpx.get(f"{base}/sessions").json()["default"]

scenario = httpx.post(
    f"{base}/sessions/{session}/scenarios",
    json={
        "tag": "my-run",
        "dataset_key": "Master data",
        "algo_name": "Slow",
        "algo_params": {"duration": 5},
    },
).json()

httpx.post(f"{base}/sessions/{session}/scenarios/{scenario['id']}/run")

while True:
    status = httpx.get(
        f"{base}/sessions/{session}/scenarios/{scenario['id']}/status"
    ).json()
    if status["status"] in ("complete", "failed"):
        break
    time.sleep(0.5)

print(httpx.get(f"{base}/sessions/{session}/scenarios/{scenario['id']}").json())

Error mapping

Response shape is always {"detail": "<message>"} — branch on the status code, not the message text.

For the full reference (including the parameter-descriptor schema and CORS configuration) see the HTTP API reference in the published documentation.