FreeBlock AI: Building an AI Co-Pilot for Project Management (Without Replacing People)

Inside FreeBlock, we’ve started building a new internal startup: FreeBlock AI.

At its core, it’s a system designed to automate and strengthen project management with AI—not in the “replace people” sense, but in a very practical way: stop losing context, move faster, and reduce the day-to-day chaos that inevitably builds up across chats, calls, task boards, and repositories.

The principle: pick the lowest-hanging apple

We’re approaching this the way I like to approach most product work: start with the lowest-hanging apple. In other words, we begin by automating areas where we can get quick, obvious value right away.

One of those areas is the work of a project manager.

One place for the entire project context

The core idea behind FreeBlock AI is simple: gather the entire project context in one place—inside a single AI chat powered by an LLM.

That context includes:

• client conversations
• team conversations
• calls with the client and the team
• specifications, prototypes, and updates/addendums
• Trello tasks and activity history
• GitHub commits and their analysis
• voice messages with transcription to text

The result is a single, queryable source of truth you can access through a Telegram bot or a web interface.

What this unlocks for the team

When all the context is collected and searchable, the day-to-day workflow becomes noticeably easier:

• a PM can prepare client reports much faster
• the team can ask questions about the project and get grounded answers
• developers can clarify details in the spec and revisit the history of decisions
• important things don’t disappear inside chat threads, calls, or task managers

Event logging: fewer tabs, less manual digging

We’re also building logging for key events—commits, task movements, and meaningful project changes. The bot can bring these updates directly into the project chat, so you don’t have to manually dig through dozens of sources just to understand what happened and when.

Under the hood: this requires real RAG, not just a long prompt

Obviously, a system like this can’t rely on “just a long context window.” To make it work reliably, we need a proper RAG pipeline with vector search and high-quality data ingestion and structuring.

But the idea itself remains very straightforward:

If AI is going to help teams deliver projects, the first thing it should do is protect meaning—agreements, decisions, and the history behind them.

This is only the beginning

FreeBlock AI is still at the start, but I’m planning to run a separate series of posts about it. I’ll share how development is going, what we’ve already added, which features we’re testing, where we hit limitations, and how we’re moving the product forward.