Inspiration
I've wanted to build apps that can automate entire workflows, and carry out full and automated tasks for users, for a while, as have others building in the AI space. Such an agent is possible now with MCP, enabling a common protocol for multiple different apps.
Also, when building AI systems before, I noticed that models on their own can't always plan out long term or real world tasks, and often fall short of completing them, which is another thing I hoped to solve with this system, by using brain-inspired algorithms like distributed processing and associative memory to promote more reliability and personalisation.
What it does
The tool can take real world actions for users on apps like Gmail, Notion, and Twitter, after offering a one-click integration to each of these apps.
The idea was to promote ease of use and convenience, allowing users to slowly automate workflows that need usage across multiple apps
How we built it
I built this agent on top of Composio, a service that enables authentication and action taking on various apps using MCP.
Specifically, I created a system using brain-inspired algorithms that calls apps through Composio, effectively providing an AI secretary for users to take real world actions on apps.
Challenges we ran into
A few challenges I ran into while building this were figuring out how to use Composio effectively to connect with multiple apps, specially with authentication handling.
I also ran into other challenges such as designing the system to take the right actions, to identify information in context, and other such issues. I handled these by implementing specific reasoning algorithms to promote reliability and also upgrading the LLM I was using to a better version (lol)
Accomplishments that we're proud of
I'm proud of the fact that the assistant can actually identify what the right app, and then, the right tool, to use is.
Also, this was my first time building an iOS app, and I'm happy with how it ended up, and the fact that it builds successfully and runs smoothly on my phone.
What's next for Saidar
I intend to expand this into a full secretary service, including handling files, setting up automations based on factors like geolocation, time, etc., building a call-based interaction, and many more expansions.
I want Saidar to be the way everyone interacts with their apps in the future, almost like an abstraction layer over the internet.
Built With
- composio
- python
- supabase
Log in or sign up for Devpost to join the conversation.