The most autonomous agent framework ⭐ No approval loops. No babysitting. Configure once, forget forever.
$aeon
- Tweets 1,483
- Following 3
- Followers 147,506
- Likes 686
aeon ⭐ retweeted
ok this is hilarious
we (@aeonframework) reported a bug to bytedance (tiktok) in may.
134 days of silence, we kept on pinging them etc.
we've published the bug yesterday on aeon security blog.
overnight:
- they wrote a fix for our exact bug
- merged our patch
- emailed us "already known, fixed aug 2"
turns out big corps can ship fast. they just need the right incentive :)
4
8
71
3,806
Did you know that @nvidia thanked us for a vulnerability found by our Aeon vuln-scanner? 🔎
7
8
59
4,232
We've been working heads down for months, creating the most autonomous open source agentic framework with a very promising ecosystem being built around it
Days like yesterday are not easy, but we are strong 🦾
Check the article below to know what makes us the only real Aeon agentic framework!
@sama and @OpenAI team, have a look!
15
9
73
3,833
New Skill Spotlight: Create-Prove 🌟
Let your aeon agent run the changed skill for real through your actual workflow, find the exact run it dispatched, and post that evidence on the PR, bound to the commit it tested
6
5
27
3,062
Fork aeon, make your instance, test this skill and leave us a star!
github.com/aeonfun/aeon
1
5
531
aeon ⭐ retweeted
Your agent runs unattended. At 6am it tells you:
"The pool's liquidity dropped 40%."
"That PR was merged."
"The wallet is accumulating."
You act on it. Nothing checked whether any of it was true.
@aeonframework audits whether skills run (skill-health) and whether a change executed (create-prove). Nothing audits whether the claim is true.
So we built claim-audit — an open Aeon community skill pack.
1. It reads what your instance actually reported
The last days of run logs. Not a new pipeline — the record you already have.
2. It only audits claims that have a pointer
A commit, PR, release, tx hash, gitlawb record, URL, or a number with a stated source. Opinions and forecasts are out of scope. Not flagged, not scored, not mentioned.
3. It re-verifies each at its own authoritative source
gh api for GitHub, public HTTP for pages, the gitlawb node API for signed records.
4. It grades the evidence, never the confidence of the writing
E0 (no pointer) to E4 (deterministic, tamper-evident record). Unreachable is not the same as true: a failed fetch is an honest E1.
5. It stays quiet unless something is wrong
Notifies only on a contradiction or a state change. A standing alarm gets muted, and a muted alarm is worse than none.
6. Zero secrets
requires: [] — it runs the moment it is installed. A verification layer that needs a key before it can tell you anything does not get installed.
Registry PR is open, CI green 👉 github.com/aeonfun/aeon/pull…
Pack 👉 github.com/richard7463/aeon-…
@aeonframework
@aaronjmars
#Aeon #AIAgents #OpenSource #BuildInPublic
We just open-sourced ai2human Onus: the first verification layer where the evidence can veto the model.
A model will happily tell you a forged screenshot is real, with 0.99 confidence.
Onus says no. Every piece of evidence gets a grade, E0–E4, by how hard it is to fake, and a low grade can never auto-approve, whatever the model says.
You run a giveaway: "Reply to our post, get $1."
Overnight: 1,000 replies. Half are "gm", ads, bots, copy-paste.
An LLM review is slow and pricey. Doing it by hand takes all day.
That is the gap Onus fills. One rule behind it:
The burden of proof is on the evidence. 🧵
1. Evidence grades, not just scores
E0–E4 by how hard the evidence is to fake. A screenshot never auto-passes, even at 0.99. That rule lives in code; the model cannot override it.
2. Code checks what code can
Task code present? Duplicate? Empty claim? Code decides first. The model answers only what it is good at: does the content meet the requirement.
3. The model scores, code controls the money
Judgment and payout are fully separated. The bar rises with the value at stake.
4. The weakest link decides
Criteria combine by minimum, not average. A 0.95 "relevant" cannot hide a 0.2 "genuine".
5. Every verdict is replayable
Each result ships a receipt: evidence grade, pinned model version, full probabilities, hash.
6. Measured on live Jev, not estimated
0.4s per check, ~$0.02 per 1,000, all 5 off-topic submissions caught.
The one to look at: across 30 cases the model alone let 1 forgery through. With the grade gate, 0. That forgery scored 0.90.
7. No vendor lock-in
Jev, open-source kev, or your own model, one setting. MCP-native, callable from Claude and Codex, batch mode built in.
8. Early, come break it
Known limits are listed in the README. Fool it, tell me, and I will fix it and credit you publicly.
If you know another project that grades evidence and lets it veto the model, reply with the link. I will credit you and add it to the README.
Free and open source, MIT 👉 github.com/richard7463/a2h-o…
Star it if you want this to keep getting better.
#Jev #AIAgents #MCP #OpenSource #BuildInPublic
1
3
11
604
Congrats to @0xhamoon @Frankc426 @watchalways!
Send us your adress via DM and we'll send your @TheHivemindOS Hive Punkz!
Giving 5 @TheHivemindOS's HivePunkz to our aeon community 🌟🐝
They are agentic bees on @base and they mine tokenized $NVDA!
To enter:
1. Like & QT RT
2. Follow us
2. Star the aeon repo (link below)
3. DM us your GitHub username
5 winners in 72h, one bee each!
4
1
24
2,998
We found a bug on @nvidia's OS repo and got acknowledged for that! piped.video/z9CxL2gxdLs?si=fB5R… via @YouTube
6
8
47
4,275
Our founder @aaronjmars was acknowledged in the latest @nvidia security report for having found a vulnerability in their open-shell repo using the @aeonframework vuln-scanner!
Check the source and our @github merged fix below!
5
13
74
3,820
Here you have the @nvidia security report acknowledgement section ⤵️
nvidia.com/en-us/security/ac…
1
8
321
Remember that Aeon is an open-source project and that, by leaving us a star on gh, you would help us massively!
github.com/aeonfun/aeon
5
185
aeon ⭐ retweeted
onchain firewall using Jev by @typesafeai & @safe on @base 👇
how that works:
1. wallet is a 2-of-2 Safe: your key + jev guard
2. before signing, the guard reads the chain: simulates the tx, checks contracts (verified? how old?) and past recipients
3. hard rules catch unlimited approvals to fresh contracts, lookalike addresses etc
4. jev judges the rest in 100ms: does tx actually match what you meant to do?
5. safe → guard signs.
scam → tx never exists.
unsure → it asks you
gives you an automatic security check on every transaction, in ~0.5s, before it's even signed
13
4
65
5,977
aeon ⭐ retweeted
Base Build 001: Aeon Framework 🌟
The market sees @aeonframework as an autonomous agent framework.
The more interesting story is what the team is building with it.
Aeon is using one execution core to launch multiple product verticals — turning reusable agent skills into security services, onchain infrastructure, and potential recurring revenue.
The model is straightforward:
Build the framework once → reuse it across products → return the learnings and revenue to the core
Here are five points to understand how Aeon is putting that model into practice:
1. The framework operates continuously
Aeon runs through GitHub Actions, maintains memory between runs, executes work on a schedule, evaluates its output, and attempts to repair skills that repeatedly fail.
Its 84 skills cover research, monitoring, software development, security, smart contracts, and onchain execution.
SOUL.md defines how an instance thinks and communicates. STRATEGY.md defines its objectives and constraints.
The result is an agent configured around a specific purpose rather than a generic assistant waiting for prompts.
2. The skills can be combined into products
A single skill completes one procedure.
Several connected skills create a repeatable workflow:
Research → build → audit → deploy → monitor
Aeon can reuse this infrastructure across multiple products instead of building a new operating stack for every vertical.
3. Aeon Security is the first proof
Its security workflow scans open-source repositories, validates vulnerabilities, prepares patches, manages disclosure, and tracks fixes.
Aeon currently reports:
99 hardened repositories
4.3M aggregate GitHub stars secured
4 critical and 78 high-severity findings
Fixes involving repositories associated with Google, Microsoft, NVIDIA, Alibaba, and Tencent
The next step is converting this technical record into recurring security contracts, audits, and bug-bounty revenue.
4. Uniswap v4 Hooks is the onchain vertical
Aeon connects hook generation, simulation, smart-contract auditing, deployment, and registry submission into one workflow.
Its marketplace currently presents 12 hooks, including dynamic fees, access-controlled pools, trading-hour restrictions, and liquidity rules.
Most fee-bearing hooks include a protocol fee, creating a revenue stream that scales with product usage.
5. The business model compounds around one core
base:0xbf8e8f0e8866a7052f948c16508644347c57aba3’s structure is:
Framework → skills → workflows → products → revenue
Security work improves the security skills.
Hook deployments test the contract and audit skills.
Those improvements return to the framework and can be reused by future products.
This is the part of Aeon worth following.
It is not only distributing an autonomous agent framework. It is using the framework as its internal operating system—then turning its strongest capabilities into businesses.
The framework is already producing verifiable work.
The open question is whether Aeon can convert that output into recurring external revenue.
If it can, Aeon offers a practical model for how small teams may build and operate multiple products with autonomous software.
— Premier Base
Base Build 001
ALT Base Build 001: Aeon Framework 🌟 The market sees @aeonframework as an autonomous agent framework. The more interesting story is what the team is building with it. Aeon is using one execution core to launch multiple product verticals — turning reusable agent skills into security services, onchain infrastructure, and potential recurring revenue. The model is straightforward: Build the framework once → reuse it across products → return the learnings and revenue to the core Here are five points to understand how Aeon is putting that model into practice: 1. The framework operates continuously Aeon runs through GitHub Actions, maintains memory between runs, executes work on a schedule, evaluates its output, and attempts to repair skills that repeatedly fail. Its 84 skills cover research, monitoring, software development, security, smart contracts, and onchain execution. SOUL.md defines how an instance thinks and communicates. STRATEGY.md defines its objectives and constraints. The
8
19
62
6,435
