Seryvon

Deterministic web audit tool across 5 pillars — SEO · GEO · GSO · AEO · ASO.

Seryvon measures a site's visibility to traditional search engines and to generative engines, AI Overviews, answer engines and — uniquely — autonomous agents (the ASO pillar). Every score is computed, traceable and reproducible: two audits of the same site yield the same result.

Product identity: Seryvon is published by Powehi (powehi.eu).

The 5 pillars

A single weighted, renormalized global score aggregates the five pillars. Criteria that cannot be measured (e.g. a missing API key) are marked not_measured and excluded from the computation — never estimated.

Project status

✅ Open core complete — Observe + Diagnose stable (through Phase 10).

What is already delivered:

Verification actuelle : 691 tests backend passant dans le profil Docker dédié, avec 100 % de couverture Python (5 163 statements, 0 ligne manquante). Le frontend compte 200 tests, avec typecheck propre et 100 % sur les lignes, branches et fonctions. The frontend uses React Router 7.18.3 and has no known production dependency vulnerability (npm audit --omit=dev).

Next: hardening of the local open core, release automation and connector maintenance.

seryvon run https://example.com      # full audit -> JSON/HTML/Markdown report
seryvon aso https://example.com      # ASO pillar only (agentic readiness)
seryvon citations https://example.com --dry-run   # LLM citation tracking (BYOK)
seryvon history example.com          # history of persisted audits

The full design documentation lives in docs/sic/ (Search Intelligence Core — the internal name for Seryvon's measurement kernel).

Quick start

With Docker (recommended)

cp .env.example .env          # adjust if needed
docker compose up --build     # postgres, redis, api, workers and frontend

The local stack binds to loopback only: frontend at http://localhost:3000, API at http://localhost:8000 (docs: /docs). PostgreSQL and Redis are also loopback-only. Migrations run automatically through the one-shot migrate service. Stop with docker compose down; add -v only when intentionally resetting local data.

Run the complete backend/PDF test suite in its dedicated Docker image:

docker compose --profile test build test
docker compose --profile test run --rm test

La suite d’intégration doit être exécutée dans ce profil : PostgreSQL utilise seryvon_test, Redis est séparé du runtime et chaque test est limité à 30 s pour rendre les blocages immédiatement visibles.

The test profile is never started by the normal runtime stack and therefore does not add development dependencies to the production API or worker images. Its image reuses the PDF runtime layer from base, so subsequent builds only install the development tools when source metadata changes.

Locally (without Docker)

python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
seryvon run https://example.com

Development

ruff check .          # lint
ruff format --check . # formatting check
mypy src              # static typing
pytest                # backend tests
cd web && npm run test:coverage -- --run

Current verification: 691 backend tests pass in Docker, 200 frontend tests pass, and the frontend reports 100% statements, branches, functions and lines. The backend coverage command reports 100% globally (5,163 statements, 0 missed). Backend tests requiring a live database should be run with SERYVON_TEST_DATABASE_URL set; PDF tests additionally require WeasyPrint's native libraries. The production Docker image includes the PDF runtime, while development test dependencies stay outside that image.

Technical performance and limits

Measurements below are local reference values, not a performance SLA:

Crawler requests are bounded, robots-aware and SSRF-validated. PDF and other non-HTML resources are not sent to Playwright. External connector calls remain dependent on provider latency, quotas and credentials.

Architecture

A Python 3.12+ core (async httpx crawl, deterministic scoring, FastAPI API, Typer CLI) plus a TypeScript/React dashboard, backed by PostgreSQL + Redis (+ optional S3/MinIO for raw artifacts), fully containerized. Scoring follows a "rule registry" pattern: each criterion is a self-registering rule, so new ones can be added without reworking the engine.

The internal design follows the Search Intelligence Core (SIC) model:

OBSERVE → DIAGNOSE → ACT → PROVE

Observe and Diagnose define the current open-core scope. Design specifications: docs/sic/.

The multi-tenant SaaS is intentionally maintained outside this repository in the private seryvon-saas repository. Core remains local-first and single-tenant; SaaS-specific authentication, isolation, quotas and continuous operations must not be added here.

License

Seryvon (the core) is distributed under the GNU AGPL-3.0-or-later license. Copyright © 2026 Powehi — https://powehi.eu

You may freely use, modify and redistribute Seryvon. If you run a modified version as a network service, the AGPL requires you to make the corresponding source code available to your users.

Reused third-party components (GEO Optimizer, OpenSEO) remain under the MIT License; see NOTICE for attributions.

Open core. This repository holds the free core (the 5 audit pillars). The operated offering (managed hosting, continuous monitoring, collaboration) is a separate proprietary product, published by Powehi.

Built with components adapted from GEO Optimizer (© Juan Camilo Auriti) and OpenSEO (© Ben Senescu), used under the MIT License.