André Koga

Founder & CTO @ Harumi | @harumi_ai_

I write my own posts, no AI used.

[André Koga @andrehkoga]

When people ask us whether we're competitors of big AI companies such as Anthropic or OpenAI, I always answer the same thing: We're working in tandem with these cutting edge models, not against;

Our platform harnesses their power, using a solid infra to let users run whatever optimization they want.

If models plateau or slow down in terms of improvements, that doesn't change our goals to become the lowest friction platform for OR and PO.

If models become strong enough to solve complex scheduling, routing or whichever kind of problem you got, then our platform will be even better at harnessing their capabilities.

Check what we're building at https://t.co/fMcRcIq4cj

@harumi_ai_

[André Koga @andrehkoga]

3/3 You can now @mention a PDF straight in chat and have it read the actual text, not just the attachment icon. Tool responses can only carry plain text, so getting there meant extracting real content from the file instead of passing it through untouched.

[André Koga @andrehkoga]

2/3 Open-ended asks ("analyze everything about X") used to run the agent unbounded until it hit a hard step limit and returned nothing. Now it takes one bounded pass, summarizes what it found, and offers to continue, narrow, or stop — no more silent runaway search.

[André Koga @andrehkoga]

Charts got a cleaner, more consistent look too, and everything now arranges itself into a sensible layout without any extra setup. The AI assistant that helps build dashboards already knows about all of it, so it won't trip over these new pieces.