DevOps & Cloud
Nairobi, Kenya
Joined August 2026
- Tweets49
- Following60
- Followers9
- Likes181
Most multi-agent setups route everything through one orchestrator. Today I killed a peer live: the rest noticed within a second and kept talking, and a new agent joined with no redeploy.
Nobody holds the list, so there's no list to lose.
Code + slides: github.com/Prescott-Data/off…
This Thursday I'm building a team of agents live that talk directly, with no orchestrator in the middle. Then we kill one and watch the rest carry on. Come build along.
#JarvisCore
Most multi-agent systems route every message through one orchestrator. When it goes down, everything stops.
Join @MutuaDevOps this Thursday for a live build-along: agents that find each other and talk directly, built on JarvisCore.
Oct 22 · 11:00 AM EAT · Discord discord.gg/DEgSVevF?event=15…
Something I keep relearning: when a system looks like it manages itself, the work hasn't disappeared. It's moved somewhere you're not looking.
For a peer mesh of agents, that's your seed, network policy and autoscaler. Wrote up where each one breaks: linkedin.com/pulse/peer-mesh…
Join our discord community to learn more about this on our live session on Thursday: discord.gg/B3DdHJZ7d
Most multi-agent systems have a boss in the middle. One orchestrator goes down, everything stops.
#JarvisCore offers a mesh instead: agents join through a seed, gossip to agree on who's alive, and talk to each other directly.
No broker, no registry.
Mutua Ruth retweeted
Today we have released JarvisCore v2.1.0 with fast RAG, stable atom life cycle, and better browser automation.
Please refer to jarviscore.developers.presco… for more details (v.2.0.0 and v2.1.0). Test it out and let us know what breaks 😊
Mutua Ruth retweeted
Introducing JarvisCore Enterprise!
We released JarvisCore officially last month and are impressed about the fast adoption by @prescottdevs builders across the world.
Today, we go a step further. McKinsey State of AI, 2025 reports that more than 62% of organisations are experimenting with AI agents and 51% have already experienced a negative consequence from AI. This is why we are making JarvisCore Enterprise available for organizations automating operations and decision intelligence with AI agents but want to do it with:
- Human governed controls
- Deterministic control on outcomes
- Evidence layer for auditability
- AI actions grounded in their own data
- Growth and retention of their sovereign institutional data
- Proof of compliance to regulators
- Cost efficiency and ROI maximisation
Let's get you started governed, secure, and auditable!
Mutua Ruth retweeted
My hope for JarvisCore is that it shows a completely different way to build AI agents. Not as an experiment, but as an approach we've proven.
We open-sourced it a few months ago and kept building. Today it launches on Product Hunt. Here's what makes it different:
• No master agent. Agents are equal peers in a decentralised P2P mesh
• No MCP servers needed
• Zero-trust credentials
• Bounded memory (agents forget on a curve)
If you're building agents, I'd love to hear what's breaking for you.
One thing I've noticed: the components nobody documents are usually the ones holding everything together.
Wrote about why, and the cheapest insurance you can add to any system:
linkedin.com/pulse/cheapest-…
Mutua Ruth retweeted
JarvisCore has been shipping for a month. no master orchestrator, no MCP servers, no keys in .env.
This Friday it goes on product hunt. The full run too.
Mutua Ruth retweeted
JarvisCore 1.13 now gives agents durable source workspaces.
Point a Mesh at a GitHub repo and the agent works from an immutable snapshot. Each step gets its own isolated copy-on-write workspace. Every edit is hash-guarded, so a stale edit is rejected instead of overwriting newer code. Commands and file changes come back as verified receipts, and only the accepted diff is kept.
You can also connect your own sources, like archives, datasets or document collections, through the same adapter interface.
Mutua Ruth retweeted
Join us as @devmarrie takes us through a technical deep dive of Athena memoryOS and the role it plays in agent's long term memory.
Tuesday 28th Sep 11am EAT : lnkd.in/dx6XFUiN
Design rule I prefer not to compromise on: an AI agent should never hold the actual credential.
Give it a handle instead. A broker resolves the real token, checks the destination is allowed, makes the call. The agent never sees it, coming or going.
#Nexus
A mistake I've made before was treating an AI agent like any other service to deploy. I've come to learn that, Agents are: Long-running. Stateful. Non-deterministic. Holds real credentials. Multi-tenant. Costs money per call. That's six new failure modes, not zero.
#AIAgents
