Duration: 55:55
🧾 Analytical Summary
A highly technical AMA with Antony Lesuisse (CTO) covering performance diagnostics, scaling architectures, upgrade paths, AI design, logging/audit approaches, rendering, and developer workflows. The throughline: keep the product simple for users while offering pragmatic, scalable patterns for complex enterprise needs.
⚙️ Performance diagnostics and logging
- Odoo request logs expose: number of SQL queries, cumulative SQL time, and total request time (deduce Python time = total − SQL). Slow requests are highlighted.
- For front-end slowdowns (e.g., product pages with many variants), v19 includes major optimizations targeting sub-100ms product page loads.
🏗️ Scaling patterns (vertical and horizontal)
- Odoo.sh can scale vertically up to ~256 workers; for higher loads use horizontal distribution: multiple Odoo app servers plus PostgreSQL master + read replicas (90% read traffic typical).
- Horizontal scaling requires shared filestore and DB replication; bottlenecks are usually DB-side.
- For very large prospects, Odoo runs case-by-case POCs reproducing critical workloads to validate SLAs before committing. Confidence and correct algorithmic choices matter more than raw CPU.
🔄 Studio to code: maintainability over time
- Heavy Odoo Studio customizations should migrate to versioned Python modules.
- Export Studio changes as module data (XML), convert server actions to Python, and use Odoo’s upgrade tooling to safely rename fields/schema (propagates to filters, etc.).
- AI can help translate boilerplate while engineers ensure correctness.
🤖 AI design: model-agnostic, safe-by-default
- AI started earlier with website content and OCR; v19 generalizes via an AI module defining agents and tools.
- Strategy: no per-customer training; use large-context LLMs with RAG and let customers choose providers (OpenAI, Google, DeepSeek, etc.). Odoo may host inference in future for data residency.
- Access rights: foreground AI acts strictly under the requesting user’s permissions; background agents are constrained to tightly scoped tools (e.g., sort to folder only).
- Hallucinations are mitigated by limiting tools and requiring human checks in sensitive flows (like accounting).
- Token policy: free on Odoo Online for now; Odoo.sh not free to avoid abuse; long‑term affordable options and own keys supported.
🔐 Audit, BI, and external replication
- Track fields/chatter provide business‑level audit trails; German accounting compliance uses this.
- On Odoo.sh, external real‑time DB replication isn’t exposed; self‑hosting can archive PostgreSQL WAL for full change capture.
- For granular data platforms, consider outside pipelines; Odoo focuses on app‑level auditability.
🧾 PDF rendering and assets
- Moving away from wkhtmltopdf is in progress; options explored include Chromium‑based rendering and dedicated HTML/CSS engines.
- Future renderer will fetch assets via sockets/pipes instead of HTTP callbacks to avoid worker contention.
- Today, ensure proper wkhtmltopdf build and system parameters; rate limit/queue long prints.
🧮 CRM probabilities and algorithms
- The lead probability explanation shown in CRM uses a classic Bayesian algorithm (not LLM) with transparent feature attributions.
🧰 Async, background work, and resilience
- Prefer simplicity over async/await for business code; use cron/queue for long tasks with user notifications.
- gevent workers exist for async patterns when needed; ensure system recovers from worker crashes; report non‑recovering Odoo.sh incidents to support.
🧩 MCP and API exposure (exploration)
- Internally, agents are granted tool access via Python definitions. Odoo is considering generating machine‑readable API descriptors (JSON) from its API docs to experiment with broader tool exposure; caution advised.
🧠 Viewpoint: Odoo Perspective
⚠️ Disclaimer: AI-generated creative perspective inspired by Odoo’s vision.
Our job is to make complex things feel simple—so SMB users stay productive, while architects still get clean paths to horizontal scale, read replicas, and safe AI agents. When a workload is critical, we don’t theorize—we reproduce it, measure it, and optimize where it matters.
The AI direction remains pragmatic: model‑agnostic, permission‑aware, and scoped to real tasks. And we’ll keep investing in fundamentals—rendering, logging, upgrades—because reliability compounds faster than hype.
🏢 Viewpoint: Competitors (Enterprise Technology)
⚠️ Disclaimer: AI-generated fictional commentary. Not an official corporate statement.
Odoo’s guidance on read‑replicas and scoped agents is sensible for mid‑market scale. Enterprise buyers will probe deeper on managed horizontal scale, external replication, and governed AI tool surfaces. The renderer transition and Odoo.sh resilience will be watched closely in high‑traffic commerce.
The Studio‑to‑code migration story is strong; success hinges on tooling depth, upgrade reliability, and first‑class observability for complex estates.
Disclaimer: This article contains AI-generated summaries and fictionalized commentaries for illustrative purposes. Viewpoints labeled as "Odoo Perspective" or "Competitors" are simulated and do not represent any real statements or positions. All product names and trademarks belong to their respective owners.