Agentia
A mobile platform for creating AI agents and connecting them into pipelines.
I built Agentia to make it possible to create and manage AI agents from a phone. An agent is configured through four steps: identity, personality, capabilities and publishing. Multiple agents can then be connected in a pipeline, with the output of one used as the input of another.
Features
- Agent creation: Create an agent through four configuration screens without writing code.
- Agent City: Agents are displayed on a map, grouped into islands by category.
- Pipelines: Connect multiple agents, share pipelines and copy and modify pipelines created by other users.
- Agent Hub: A section for publishing and rating agents, with usage-based credits.
- Tools: Eight tools including web search, summarisation, URL reading and code execution. Agent execution uses LangChain and Gemini.
- Webhooks: Pipelines can be triggered by external systems using a private token.
Technical setup
| Layer | What I used |
|---|---|
| Mobile | Flutter, Riverpod, go_router |
| Core API | Java 17, Spring Boot 3, JPA, Flyway, JWT |
| AI worker | Python, FastAPI, Celery, LangChain |
| Web dashboard | Next.js |
| Data | PostgreSQL 16, Redis 7 |
| Infrastructure | Docker Compose, Kubernetes, GitHub Actions |
| Monitoring | Prometheus, Grafana, k6 |
Infrastructure took up a large part of the project. I first ran all ten services together with Docker Compose and later moved them to Kubernetes. GitHub Actions runs tests, linting and image builds on every push.
I created Grafana dashboards for monitoring and used k6 to run load tests with up to 100 concurrent users.