iiinova — where your digital self keeps evolving

Martin Krasser

@krasserm.bsky.social

Freelance ML and AI engineer, distributed systems developer, martin-krasser.com

Martin Krasser @krasserm.bsky.social · 07/08/2025

The key is a clean separation of concerns: your agents focus on reasoning and tool use 🧠, while a service like Composio handles API integration, authorization, and MCP server hosting for you 🔧 blog post: t.co/qPpGVROEgD

t.co

https://gradion-ai.github.io/agents-nanny/2025/08/06/agent-authorization-without-the-pain/

011

Martin Krasser @krasserm.bsky.social · 07/08/2025

Agent authorization and API integrations can be a challenge 🤔. Our latest blog post explores how you can authorize your agents to access 250+ APIs (Gmail, Slack, Notion, ...) with just a few lines of code, letting you focus on core logic, not plumbing and OAuth details 👇

111

Martin Krasser @krasserm.bsky.social · 25/07/2025

Agents have the full context of a conversation and act on behalf of users. Build your own agents in minutes with Hybrid Group's agent builder. Find out more at github.com/gradion-ai/h...

011

Martin Krasser @krasserm.bsky.social · 25/07/2025

Hybrid Groups is agentic AI for human teams. Agents understand and contribute to conversations on Slack and GitHub. Watch the demo 🎥 www.youtube.com/watch?v=OxOm...

youtube.com

Hybrid Groups

YouTube video by Gradion AI

123

Reposted by Martin Krasser

Manuel Bernhardt @manuelbernhardt.bsky.social · 20/06/2025

Multi-user, multi-agent platform: github.com/gradion-ai/h...

github.com

GitHub - gradion-ai/hybrid-groups: A multi-user, multi-agent group collaboration platform and library

A multi-user, multi-agent group collaboration platform and library - gradion-ai/hybrid-groups

011

Martin Krasser @krasserm.bsky.social · 06/02/2025

Blog post: krasserm.github.io/2025/02/05/d... Repository: github.com/gradion-ai/f...

krasserm.github.io

Exploring DeepSeek-R1's agentic capabilities through code actions - Martin Krasser's Blog

000

Martin Krasser @krasserm.bsky.social · 06/02/2025

New blog post: Exploring DeepSeek-R1's agentic capabilities through code actions (link below). DeepSeek-R1 doesn't support tool use natively, but can be used for agentic tasks through code actions. The post also covers limitations like long reasoning traces and "underthinking".

100

Martin Krasser @krasserm.bsky.social · 04/02/2025

Additional RL on agentic tool use via code actions should further improve performance. Find evaluation details at github.com/gradion-ai/f....

github.com

000

Martin Krasser @krasserm.bsky.social · 04/02/2025

DeepSeek R1 surpasses Claude 3.5 Sonnet on agentic tasks even if it wasn't trained for it. In our experiments, we leveraged DeepSeek-R1's strong coding abilities to act via code and initial results look promising.

100

Martin Krasser @krasserm.bsky.social · 30/12/2024

Learn how freeact agents work: Docs: gradion-ai.github.io/freeact/ Repo: github.com/gradion-ai/f...

000

Martin Krasser @krasserm.bsky.social · 30/12/2024

Freeact agents write their actions in code, rather than just being agents for writing code. This design provides a vast action, space—no JSON tool configs required.

100

Martin Krasser @krasserm.bsky.social · 30/12/2024

I just released freeact - a lightweight Python implementation of AI agents that use code actions to dynamically interact with and adapt to their environment.

High-level architecture of a freeact agent system

110