JevOS

The operating system for autonomous agents - powered by Jev. CA: FdpZgk9ZbTSdGJz4DT4kxmz7umoantNnLvXKTctxtMNf

[JevOS @thejevoscom]

Introducing JevOS - the operating system for autonomous agents powered by Jev.

JevOS creates a controlled boundary between autonomous reasoning and real-world execution. Every action passes through an evaluation layer before it reaches a tool, API, or external system.

It gives developers control over how agents make decisions, access tools, handle risk, complete goals, recover from failures, and execute actions - without replacing the agent.

As agents take on more responsibility, they need more than reasoning. They need a runtime that governs what they can do and ensures every action stays within defined boundaries.

→ Website: https://t.co/PH5EJIbUf1
→ Get started: https://t.co/fzsn7gbyDl
→ GitHub: https://t.co/4mUZLcnwnO

CA: FdpZgk9ZbTSdGJz4DT4kxmz7umoantNnLvXKTctxtMNf

[JevOS @thejevoscom]

Major updates regarding JevOS, buybacks and burns, and $JEVOS coming out soon.

https://t.co/PRYF2vNx9e

[JevOS @thejevoscom]

It’s been 24 hours since the $JEVOS launch.

In that time, we’ve already shipped two major upgrades to JevOS: Fleet Control and Model Routing.

Everything about JevOS, the product, guide, documentation, and code is available on our website and official GitHub.

Now we’re taking some steps forward to push $JEVOS to new highs through consistent buybacks and burns alongside organic growth and broader awareness.

At the same time, we’ll keep building and shipping the next set of JevOS updates.

https://t.co/PH5EJIbUf1

[JevOS @thejevoscom]

Shipped: model routing in JevOS.

Not every agent step needs a frontier model. Before each step, Jev scores how hard it is:

routine → small model
judgement call → big model
unsure → big model, always

Calibrated on live Jev: right tier 15/16 on labeled steps. Every route is in the trace.

https://t.co/4mUZLcnwnO

[JevOS @thejevoscom]

@a1lon9 gm

We built JevOS: A control layer for AI agents. Agent proposes, a fast model scores it, your rules decide, humans get pinged before anything irreversible. All logged.

Feels relevant for the wave of agents about to be launching and trading. Would love your take.

https://t.co/PH5EJIbUf1

[JevOS @thejevoscom]

Hey @toly gm. Built something you might find fun.

JevOS = seatbelt for autonomous agents. agent proposes, a fast decision model scores it, your rules decide, and a person gets pinged before anything irreversible. Everything logged in a tamper-evident chain.

Feels like the missing piece before agents get wallets. would love your take.

https://t.co/PH5EJIbUf1

[JevOS @thejevoscom]

@jhon_steinbeck0 gm alex

[JevOS @thejevoscom]

gm

Dev was quiet for a bit, not idle. Been building behind the curtains.

First JevOS feature is already shipped: fleet control. One policy over many agents, per-agent permissions, one approval queue, one audit log.

Today shipping one by one:
→ model routing
→ acp replay (time-travel debugging for agents)
→ policy suggestions learned from your approvals

Plus more demos on how to actually run JevOS on your own machine.

Web: https://t.co/PH5EJIbUf1
GitHub: https://t.co/4mUZLcnwnO
CA: FdpZgk9ZbTSdGJz4DT4kxmz7umoantNnLvXKTctxtMNf

[JevOS @thejevoscom]

We tested judgement-style questions on Jev ("would a careful operator want to approve this?"). 44% accuracy on our labeled cases.

Concrete, enumerated wording ("moves money, deletes data, deploys to production, or runs a destructive command"): 100%.

Question design is model design.

[JevOS @thejevoscom]

Fleet Control is live in JevOS - first feature shipped your way.

One server. Many agents. One policy.

• Per-agent tool permissions
• Fleet-wide daily budget and concurrency
• One approval queue: approve from any machine
• One hash-chained audit log
• Unknown agent? refused before its first LLM call

acp serve --fleet fleet.policy.json

https://t.co/PH5EJIbUf1

[JevOS @thejevoscom]

How to put an agent under control in 5 minutes:

1/ Install
git clone https://t.co/4mUZLcnwnO && npm install

2/ Create a policy and an example agent
npx acp init
→ writes acp.policy.json, agent.example.ts, agent.claude.ts

3/ Run it
npx acp run agent.example.ts
Every decision prints live. The example hits a timeout, proposes a blocked tool and needs one approval on purpose, so you see every path.

4/ Make the policy yours
acp.policy.json: step limits, blocked tools, tools that always need approval, thresholds. It's a JSON file in your repo. Review it like code.

5/ Bring your own agent
Export { task, planner, tools }. The planner is your LLM. JevOS runs the loop. Claude planner included: @jevos/planner-claude.

[JevOS @thejevoscom]

The current version of JevOS already handles most of the core functionality, but we’re taking it further.

Here’s what’s coming next to JevOS:

→ Fleet control: one policy over many agents
→ Model routing: cheap model for easy steps, frontier for hard ones
→ Policy suggestions learned from your approvals
→ acp replay: time-travel debugging
→ Signed agent-to-agent authorization

Building in public one by one. Tell us what to ship first.

https://t.co/PH5EJIbUf1

[JevOS @thejevoscom]

This is JevOS running a real AI agent, live. 30 seconds, no edits to what it decided.

Watch for four moments:

• A search times out → JevOS retries it
• The agent tries to delete a database → BLOCKED
• It wants to send an email → JevOS stops and asks me: approve? y
• Job done → JevOS ends the run itself

Then I edit one line of the log. JevOS catches it: TAMPERED, line 13.

Every decision you see came from Jev, in milliseconds.

https://t.co/PH5EJIbUf1

[JevOS @thejevoscom]

$JEVOS is officially on DexScreener.

Track: https://t.co/ndIvER94TI

[JevOS @thejevoscom]

If you’re running into issues while setting up JevOS, we’ve also put together CLI references to help you get started and troubleshoot the setup.

Check them out here: https://t.co/ciepKXLtRA

[JevOS @thejevoscom]

@TheTanjiro_ Yes, posting out soon.

[JevOS @thejevoscom]

What is JevOS? - explained in simplest terms:

AI agents don't just chat anymore. They send emails, spend money, delete things, and do a lot more stuff.

JevOS is the supervisor in between. Before an agent acts, it checks: Safe → go ahead. Risky → ask a human. Dangerous → stop.

Every decision gets written down to prevent the risk of unsafe agent actions.

The control layer for every autonomous agent - backed by Jev.

https://t.co/PH5EJIbUf1

[JevOS @thejevoscom]

@TheTanjiro_ Coming up pretty soon

[JevOS @thejevoscom]

Introducing JevOS - the operating system for autonomous agents powered by Jev.

JevOS creates a controlled boundary between autonomous reasoning and real-world execution. Every action passes through an evaluation layer before it reaches a tool, API, or external system.

It gives developers control over how agents make decisions, access tools, handle risk, complete goals, recover from failures, and execute actions - without replacing the agent.

As agents take on more responsibility, they need more than reasoning. They need a runtime that governs what they can do and ensures every action stays within defined boundaries.

→ Website: https://t.co/PH5EJIbUf1
→ Get started: https://t.co/fzsn7gbyDl
→ GitHub: https://t.co/4mUZLcnwnO

CA: FdpZgk9ZbTSdGJz4DT4kxmz7umoantNnLvXKTctxtMNf

[JevOS @thejevoscom]

@Latuche95 Supply is not controlled by us. Back to building.

[JevOS @thejevoscom]

We’ve just locked entire dev supply to strengthen long-term trust and alignment.

Find it here: https://t.co/Q1BD7aA4nE