iiinova — where your digital self keeps evolving

Liviu @liviu.ovh · 26/08/2026

v0.5.0 added the looptroop CLI: setup, start, stop, restart, status, open, logs, doctor, and clean. start detaches, so LoopTroop keeps running after the shell closes.

100 1

Liviu @liviu.ovh · 07/08/2026

Hey guys. If you want free #opus5 or #gpt56sol via API just sign up using this link - 200$ in tokens - agentrouter.org/register?aff... Only strict requirement is to sign up using a github account older than 2025! I've used more than 500$ in the past. Only downside - it is really slow :)

000 0

Liviu @liviu.ovh · 29/07/2026

Just hit 100 ⭐ on GitHub for LoopTroop! Local AI coding orchestrator that turns tickets into real PRs - LLM councils plan it, Ralph loops perfect it, OpenCode worktrees ship it. Huge thanks to everyone who’s starred or given feedback 🙏 #LoopTroop #opensource #AgenticAI

130 3

Liviu @liviu.ovh · 02/07/2026

9/ The tradeoff is very deliberate: LoopTroop is slow and precise. It is not meant for tiny edits you want in 20 seconds. It is for bigger multi-file work where the planning quality, context boundaries, retries, and review trail matter more than instant output.

100 1

Liviu @liviu.ovh · 02/07/2026

8/ The GUI is a modern Kanban-style board, not an invisible terminal daemon. You can watch tickets move through phases, inspect artifacts, review logs, approve planning gates, inspect bead progress, and review the final result before merging.

100 1

Liviu @liviu.ovh · 02/07/2026

7/ Execution uses Ralph-style loops. If a bead fails, LoopTroop does not keep dragging the same polluted chat forward. It records a compact failure note, resets the bead workspace, starts a fresh session, and retries with clean context plus the useful lesson from the failure.

100 1

Liviu @liviu.ovh · 02/07/2026

6/ After that, the PRD becomes the source of truth. Then the PRD is split into beads: small implementation units with clear objectives, dependencies, target files, acceptance criteria, and test commands. The agent works bead by bead instead of trying to swallow the whole feature at once.

100 1

Liviu @liviu.ovh · 02/07/2026

4/ The planning part uses an LLM council. Multiple configured models draft independently, vote on anonymized outputs, refine the winner, and run coverage checks before the next phase. This is used for the interview, the PRD, and the bead blueprint.

100 1

Liviu @liviu.ovh · 02/07/2026

3/ LoopTroop treats that as a systems problem, not just a prompting problem. Instead of one giant conversation, a ticket moves through explicit phases: repo scan -> interview -> PRD -> bead planning -> execution -> recovery loops -> final review / PR.

100 1

Liviu @liviu.ovh · 02/07/2026

1/ 🧵 I built LoopTroop: a local open-source AI coding orchestrator for people who care more about correctness than speed. It is my attempt to fight context rot with context engineering: smaller phases, durable artifacts, human approvals, LLM council planning, and Ralph-style retry loops.

110 1

Liviu @liviu.ovh · 01/07/2026

Uploaded a 2.5 min walkthrough of LoopTroop, my open source AI coding orchestrator. It turns work into a ticket flow: interview -> PRD -> beads -> OpenCode execution, with LLM council planning and Ralph-style retries. www.looptroop.ovh #BuildInPublic #ContextEngineering #AICoding

140 4

Liviu @liviu.ovh · 08/01/2025

Lol, #microsoft #copilot doesn't consider #Azure reliable (the source for his LLM)

000 0

Liviu @liviu.ovh · 07/01/2025

And the best self-service developer portal or #InternalDeveloperPortal (IDP) I have ever seen. It must be easy being a developer at @fujitsu. Great catalog

000 0

Liviu @liviu.ovh · 07/01/2025

Implementation using #HashiCorp AFT (account factory for terraform) - haven't even heard about it until now. Other considerations.

100 1

Liviu @liviu.ovh · 07/01/2025

Some trends including from #DORA #stateofdevops.

110 1

Liviu @liviu.ovh · 07/01/2025

What is platform engineering, the components, principles, and how it works in practice.

100 1

Liviu @liviu.ovh · 05/01/2025

Lol #grok opinion about me on #twitter when clicking on the #grokai button on my profile. Except for my critique about tech devices usability in Europe, everything is correct!

000 0

Liviu @liviu.ovh · 05/01/2025

The thing that I liked the most is a new definition for blameless postmortem or how they called it at #aws: COE, Correction of error - Root cause is never the fault of the operator, the system wasn't configured correctly to stop/warn him.

000 0

Liviu @liviu.ovh · 26/12/2024

Another great take: Failure is inevitable. However that doesn't mean users should know about it! And some slides about cell-based architecture and gray failures.

000 0

Liviu @liviu.ovh · 26/12/2024

#Reliability is the property of the worst system on its worst day when you needed the most under load!

100 1

Liviu @liviu.ovh · 26/12/2024

And the best definition of #reliability -with a great example when a fire broke out and no one could call the fire department as the telephone system was down

100 1