AI consulting · Small & medium businesses · US & EU
AI tools are everywhere. Working systems are rare. Odysail Systems audits, builds, and runs them for small and medium businesses in the US and Europe, inside the software you already use.
Figures from a live deployment, not projections.
What we do
Every engagement follows the same three steps: no exceptions, no surprises.
We map your data, your workflows, and your failure points before anyone writes code. You get a clear plan: what to automate, what to leave manual, and what it costs. Scoped and priced for a small business, not an enterprise program.
Custom AI agents built on Anthropic's Claude, wired into the software you already run: ERP, CRM, email, spreadsheets. Gated behind your existing company logins, access mapped to each role, shipped with automated tests. If it can't survive a hostile question, it doesn't ship.
Models change, APIs drift, teams turn over. We stay on: monitoring, upgrades, new tools as your business grows. AI is not a handover. It is an ongoing operation, and we run it with you. The goal is giving your people their hours back, not replacing them.
Method
It's not a model problem. It's an engineering problem. The models are good enough; the systems around them usually aren't.
We come from commercial sailing, where equipment either works or people get hurt. We build AI with the same assumption: instrumented, tested, accountable.
One engineer, Malek M'Sadek, owns your deployment end to end. The person on your first call writes the code, tests it, and answers when something breaks. No account managers, no handoffs, no theater.
Straight answers
It stays in your systems. We build inside your accounts, not ours, and you approve what the system can see before we build it.
Every build ships with automated tests that catch drift when models or data change. Your people stay in control: the system drafts and answers, humans approve what matters.
The audit names the hours each workflow should save before anything gets built. If AI is the wrong tool for the job, the report says so, and you keep it.
Platform
We deploy on Anthropic's Claude: one platform, known deeply, instead of every model, known barely.
Frontier models in production: prompt architecture, tool use, structured output, and cost control tuned per workload.
MCP servers that connect Claude to the systems a business already runs. Built, secured, and running in production today.
Autonomous agents and internal tooling on Anthropic's Agent SDK, the stack Anthropic itself builds with.
Automated test suites against live model behavior. Deployments ship with evals. 755 tests guard the deployment below.
Claude API · MCP servers · Agent SDK · Claude Code · Automated evals
Deployment log · 2026
A distribution business ran its whole operation through a legacy inventory ERP nobody could query without training. We gave it a language interface: Claude, connected through an MCP server exposing 43 tools over OAuth, with roles mapped to the org chart so each person sees only their slice of the business.
It reads the ERP and never writes to it. Nothing on the client's network is exposed to the internet. 755 automated tests stand between the model and the data. Executives now ask the system questions in plain English and get numbers, not guesses.
For deployment inquiries
Tell us what you run and what keeps breaking. Two sentences is enough. We'll tell you, in plain terms, whether AI helps, what it costs, and how long it takes. We work with clients across the US and Europe. The reply comes from the engineer who would build it, not a sales team.