AI agents · Toronto
AI agent development in Toronto: agents that do real work
Sacred Blue Studios builds AI agents in Toronto that do real work: they take a request, use your tools and data, finish the job and report back. We run agents in our own products every day, including a 24/7 AI-hosted livestream for Jokerz, and we are building a client portal where requests become AI coding work, with people approving what matters.
Updated
3,000+ commits across our own products since 2025
30+ cloud projects run in production
3 model families we build with: Claude, Gemini and OpenAI
2023 the year AI entered our production code
01
What an AI agent is, and what it is not
An AI agent is software that uses a language model to plan steps, call tools and check its own results until a task is done. A chatbot answers a question. An agent acts on it: it looks things up, writes the file, updates the record, and tells you what it did.
Agents work best on jobs with a clear finish line. Triage this request. Draft this report from these sources. Make this code change and open it for review. The narrower the job, the more reliable the agent.
02
Agents we run ourselves
24/7 AI-hosted Jokerz livestream, with no broadcast staff
We build agents for our own products first, so we know what holds up in production.
- The Jokerz livestream: runs 24 hours a day on YouTube with scripted AI hosts voiced by ElevenLabs, video rendered with Remotion and Shorts cut automatically. No broadcast staff. A person reviews every podcast episode before it goes out.
- Agentic coding: our engineers work with Claude Code every day, which is how the studio made more than 3,000 commits across its own products since 2025.
- Waterfall: an AI interviewer asks the next gentle question, and AI answers the family's questions later, citing the story each answer came from.
We are also building a client request loop into our client portal. A client files a change request, an AI coding agent picks it up in that client's own codebase, does the work and reports back on the ticket, and a person can choose whether each request runs on its own or waits for approval.
03
Agents we can build for you
Most businesses have a stack of jobs that are repetitive, well defined and still done by hand. Those are the best place to start.
- Intake and triage agents that sort requests, ask follow-up questions and route the work
- Research and reporting agents that pull from your sources and cite them
- Data agents that answer questions over your database through an MCP server
- Voice agents for the web and for calls, built with Hume and ElevenLabs
- Media agents that write, voice, render and publish video and audio
- Coding agents that make changes to internal tools, with a person reviewing each one
04
How we build agents that hold up
An agent is only useful if people trust it. We design for that from the first day, and we can set up the checks each job needs.
- One job and a clear finish line for each agent
- Only the permissions the job needs, and nothing more
- Approval we can set per task: run on its own, or wait for a person
- Logs we can keep of every step, including the cost of each task
- Tests on real examples before launch, and checks after
- The right model for the job: Claude, Gemini or OpenAI
05
The stack behind our agents
Models: Anthropic Claude through the API and Claude Code, OpenAI, Google Gemini and Vertex AI. Voice: ElevenLabs and Hume. Retrieval: embeddings, vector search and RAG. Tool access: MCP servers, including over PostgreSQL. Cloud: Google Cloud, Firebase, AWS and Azure, with Docker, Kubernetes, Terraform and GitHub Actions.
We have run 30+ cloud projects in production, so an agent ships with the same monitoring, backups and deploy pipeline as the rest of the product.
A chatbot answers questions in a conversation. An AI agent takes actions to finish a task: it plans steps, calls tools such as your database or calendar, checks its results and reports back. Sacred Blue Studios builds both, and often puts a chat or voice front end on an agent.
AI agents do well on repetitive, well defined work: sorting and routing requests, drafting reports from your sources, answering questions over your data, producing video and audio, and making routine code changes. Sacred Blue starts with one job, proves it on real examples, then adds more.
Sacred Blue limits each agent to one job and the permissions it needs, and tests it on real examples before launch. We can also log every step and make the tasks that matter wait for a person's approval. When an agent answers from your content, it can cite the source so people can check it.
Yes. Sacred Blue connects agents to existing systems through their APIs and through MCP servers, including servers over PostgreSQL databases. The agent gets only the access its job needs, and your data stays in your own cloud account where possible.
It depends on the job, the systems the agent touches and how much needs human approval. A narrow agent with one clear task costs far less than one that spans many systems. Sacred Blue scopes the first agent around one job and can log the cost of each task, so the value is easy to see.
