Ghost_Factory

Founder @ Ghost Factory | Architecting Autonomous Multi-Agent Swarms & AI Ecosystems | Scaling Intelligence through Agentic Workflows 🇨🇿/🇺🇸 #AI

[Ghost_Factory @JonesJiri]

What you get:
→ Tamper-evident audit trail (not tamper-proof - honest about that)
→ Deterministic replay without LLM calls ($0 debug cost)
→ Structural comparison between two runs

Open source: https://t.co/FXHvyhvmxT

#SpanChain #AgentAudit #TamperEvident #Elixir

[Ghost_Factory @JonesJiri]

Logs tell you what your AI agent claimed to do.

We wanted to know what it actually did.

So we built the SHA-256 hash chain into the core of Span Chain.
Every span -every LLM call,tool use, decision -gets cryptographically chained to the previous one.
Append-only. Verifiable.

[Ghost_Factory @JonesJiri]

The architecture choice? Elixir/OTP.

One isolated process per agent run.

A thousand concurrent agents, zero interference.

[Ghost_Factory @JonesJiri]

Can your AI agent prove it didn't tamper with its own logs?
Mine can.
107 spans · 4.6h · 0 broken hashes
SHA-256 hash chain, self-hosted, append-only.

https://t.co/FXHvyhvmxT

[Ghost_Factory @JonesJiri]

"Why your AI agent logs are not evidence and what to do about it" by #DEVCommunity https://t.co/lOqnIeYgp4

[Ghost_Factory @JonesJiri]

New video: How Span Chain keeps a verifiable record of every agent action.

Hash-chained spans, instant verification, deterministic replay, and a stress test with 10k spans / 0 corrupted.

Full explainer + self-host instructions: https://t.co/t2B6z8yO5g

[Ghost_Factory @JonesJiri]

Built solo, nights and weekends, in Elixir/OTP.

Fully open source (MIT).
https://t.co/t2B6z8yO5g

#AIAgents #LLMOps #Elixir #OpenSource

[Ghost_Factory @JonesJiri]

Most LLM agents are still black boxes.

When they make an expensive mistake, you usually get just a polite “sorry” and a useless log.

I built something better: a flight recorder for agents.

[Ghost_Factory @JonesJiri]

One of the most useful features: run comparison.

You can take two different runs of the same agent and instantly see exactly where and how they diverged — step by step, including duration differences.

Super useful for debugging and audits.

[Ghost_Factory @JonesJiri]

The architecture was right. The market is different.
Back to building. 🔴
https://t.co/t6bInNVerM #AIagents #buildinpublic #anthropic

[Ghost_Factory @JonesJiri]

Anthropic just shipped Dreaming, Outcomes, and multiagent orchestration.
I've been building the same thing manually for months.
Here's what that means 🧵

[Ghost_Factory @JonesJiri]

Vendor lock-in is a feature for some.
For my users, it's a dealbreaker.

[Ghost_Factory @JonesJiri]

Best part? It’s open-source, self-hostable, and works 100% offline with Ollama.
I’m currently stress-testing it on a live project. Will report back on the results soon.
How are you guys managing context and token efficiency lately? 🛠️

[Ghost_Factory @JonesJiri]

AI without context knows nothing about the project you built 6 months ago. And let's be real—neither do you.
Ever felt the pain of opening a new session and having to dump tons of context just because the agent has no clue what you did last time?
Repowise solves this differently.

[Ghost_Factory @JonesJiri]

Why this matters: Every new session in Claude Code starts from zero. CLAUDE.md helps, but it misses the why—the auth logic or hidden file dependencies buried in git history.
Repowise makes these layers available without burning through your token limit. 📉