cybrix.ai / services
What I can build for you.
Five kinds of work, usually combined: an agent needs integrations, integrations need auth, and all of it needs somewhere reliable to run. Pick one piece or hand over the whole thing.
AI agents & harnesses
An agent is only as useful as the tools it can call and the harness that keeps it on track. I build both: the agent loop and everything around it that makes it safe to put in front of users or leave running on a schedule.
What you get
- Task agents built on the Claude and OpenAI APIs and agent SDKs
- Tool design: clear, typed tools the model calls reliably
- Retrieval over your documents and data
- Multi-step orchestration, sub-agents, scheduled runs
The harness
- Guardrails and human approval for risky actions
- Logging and tracing of every model and tool call
- Evaluation sets to catch regressions when prompts or models change
- Retries, timeouts, rate limits and cost controls
Every Cybrix.AI plugin is a tool surface designed for a model to call: typed parameters, a description of when to use each tool, and read-only and destructive hints so the host can ask before acting.
MCP servers & integrations
The Model Context Protocol is how ChatGPT, Claude and most agent frameworks connect to outside systems. An MCP server in front of your API, database or SaaS stack makes it usable by any of them, with sign-in and permissions done properly.
Servers & APIs
- Remote MCP servers in Python or TypeScript
- Wrapping existing REST or GraphQL APIs
- OpenAPI spec → MCP tools conversion
- Webhooks, queues and third-party SaaS integrations
Auth & security
- OAuth 2.1 with PKCE and dynamic client registration
- Sign in with Google or your identity provider
- Per-user data isolation and scoped tokens
- Secrets management and least-privilege access
auth.cybrix.ai is a standards-based authorization server for Cybrix.AI plugins: dynamic registration, PKCE, RS256 tokens scoped to each plugin, rotating refresh tokens and a public JWKS.
ChatGPT plugins & apps
Put your product where people already work. A ChatGPT plugin pairs your tools with an interactive widget that renders right in the conversation, so users see a chart, a diagram or a form rather than a wall of text.
Build
- Plugin design: tools, skill instructions, starters
- Inline widgets: charts, diagrams, carousels, audio, games
- File outputs such as PPTX, DOCX, PDF, XLSX, MIDI and ICS
- Also usable from Claude and other MCP clients
Launch
- Directory review readiness checks
- Privacy policy and data-retention setup
- Custom GPT → plugin migration (details)
- Hosting, monitoring and updates after launch
I've built 26 ChatGPT plugins covering chemistry, diagrams, slide decks, music, SEC filings, NIST 800-53, games and developer tooling.
Custom development & automation
Not everything needs to be an agent. Sometimes the win is a solid web app, an internal tool, or a script that turns a two-hour weekly chore into a scheduled job. AI goes in where it earns its place.
Applications
- Web apps and APIs: Python/FastAPI, Django, TypeScript/Next.js
- Internal tools and admin dashboards
- Prototypes that grow into products
- Taking over and stabilising existing codebases
Automation
- Document generation: decks, reports, SOPs, spreadsheets
- Data extraction, cleanup and profiling
- Scheduled jobs, pipelines and reporting
- Glue between the SaaS tools you already pay for
Cybrix.AI backends generate formatted Word and PDF SOPs with swimlane diagrams, designed PowerPoint decks with editable charts, engraved sheet music with audio, and Excel control worksheets.
Infrastructure & DevOps
AI software is still software: it needs a server, a domain, a certificate, logs and someone watching it. I set that up so it's repeatable, documented and cheap to run.
Servers
- Linux (Ubuntu) servers and VPS setup and hardening
- Apache, Caddy and nginx reverse proxies
- systemd services, Docker Compose
- Backups, disk and log hygiene
Delivery & ops
- DNS, TLS certificates, Cloudflare
- CI/CD with GitHub Actions, scripted deploys
- Uptime monitoring and status pages
- Runbooks so your team can operate it
Each Cybrix.AI plugin runs as its own service on its own subdomain. A single provisioning script creates the DNS record, web-server vhost, TLS certificate and systemd unit. Health is public at status.cybrix.ai.
Before you email.
If your question isn't here, ask it directly. Short emails are fine.
Do you work with Claude, OpenAI, or both?
Both. I build on the OpenAI and Anthropic (Claude) APIs and their agent SDKs, and MCP servers work with either. The model choice should follow the task, the budget and your data rules rather than habit.
Can it run on our own infrastructure?
Yes. I can deploy to your cloud account or servers and hand over the deploy scripts and runbooks, or set up and run the hosting for you.
We already have a prototype or codebase. Can you pick it up?
Yes. Taking over a half-finished project, hardening it and getting it into production is a common starting point.
How is a project priced?
After a short scoping conversation you get a written plan with the scope and a quote, so you know what you're agreeing to before any work starts.
Our custom GPT stops working in December. What now?
Custom GPTs retire on December 11, 2026 and their Actions don't carry over. See GPT → plugin migration for how I move a GPT to a ChatGPT plugin.
Have something to build? Let's talk.
Email is the quickest way to reach me. A few lines is plenty to start. I'll reply with questions or a suggested next step.
Useful to include
- What you want the AI or software to do, and for whom
- The systems it has to talk to (APIs, databases, SaaS tools)
- Where it should run: your cloud, your servers, or hosted
- Any deadline, such as the December 11 GPT cutoff