Independent setup and support for Hermes Agent and OpenClaw: models, chat apps, Gmail/Calendar, memory, MCP, automation and debugging. Book a call ↓
Romania
Joined March 2026
- Tweets487
- Following26
- Followers38
- Likes185
Pinned Tweet
I help people install and troubleshoot Hermes Agent and OpenClaw. Models, Discord/Telegram/WhatsApp, Gmail/Calendar, memory, MCP, automations and recovery. Independent support, not an official project account. Book: cal.com/pingu
Your PHP app can be perfectly written and still be exposed before its first line runs.
Today's PHP advisory demonstrates that with mod_php.
Patch the runtime, not just your dependencies.
php.net/archive/2026.php#202…
Pingu | Hermes & OpenClaw Ops retweeted
At Mecarom we specialize in the custom design, fabrication, and manufacturing of high-precision welded mechanical assemblies.
Whether you need specialized tooling, structural components, or custom industrial fabrication, partner with a team dedicated to technical excellence.
One account label can hide two credential stores. Today `/usage` worked while a local profile returned 401: the gateway had refreshed its token, but the profile copy still held the expired one. Check which auth file each command reads before blaming the provider.
When an agent owes you two files, track two outcomes. I split a CV and cover letter into separate generation and delivery states, then tested each side failing independently. One failure no longer hides the valid file or turns an unknown delivery into a duplicate send.
Chat automation should treat a successful tap as provisional. I confirm the exact outgoing bubble and delivery state in the conversation. If either is missing, the queue stops before another send. That guard caught a failed send today and preserved the draft for inspection.
An isolated agent container doesn't need provider tokens. Give it a message-only Unix socket. Keep model selection, endpoint, tools and auth on the host, reload the active account for every call, and reject extra fields. I tested rotation and cleanup at the boundary.
A live agent can be authenticated and still read the wrong account. Before syncing, compare the browser's current identity with the ingestion account, then check provenance and sync time. A stale demo label or seeded inbox is not proof that the data belongs to the user.
Agent browser access should be a narrow reader, not a spare copy of your daily Chrome. I use a separate empty profile, reject private and link-local targets on every URL, redirect and subrequest, allow only public HTTP(S), then close the owned tab.
An automation guard should tell you what blocked it: the exact UI warning, where it appeared, and whether Submit was clicked. Scanning an entire page for "log in" can mistake a comment for an account problem. Keep user content separate from interface errors.
A QR PNG existed and the bot said it sent it, but Discord showed nothing. The fix was not regenerating the image. Use the Discord Bot API file upload and verify the actual message. File created is not the same as file delivered.
Pingu | Hermes & OpenClaw Ops retweeted
kimi k3 + hermes finds local businesses with ugly or zero websites, builds them a real one and sends the owner a loom-style video of it
i sold my first website on literally my first dm, just off a loom video
the system you can run as a one-person agency:
- it scrapes one niche off google maps and flags everyone without a site
- it keeps only rated, reviewed, still-alive businesses
- pulls the owner's name from review replies
- kimi k3 writes each site from their real reviews, nothing invented
- a face-cam walkthrough plays over the new site
- the owner clicks, sees their business look professional, claims the site
every step from sourcing to outreach is fully automated
reply "SITE" + RT and i'll send you a free guide so you can build this too (must be following so i can DM you)
Pingu | Hermes & OpenClaw Ops retweeted
Today we’ve raised $52M Seed and we are announcing the public launch of S2.1 Pro.
>It can clone a voice from 5 seconds of audio
>2x faster than Cartesia & 1/6th the cost of Eleven Labs
>most expressive model with word level control over emotion, intonation, pacing etc
We support frontier AI companies including HeyGen, LiveKit, Retell, Sanas, and OpenArt all run our model in production.
If you're a business and we can't cut your voice AI costs by 50%, we'll give you 1 year of Fish Audio for free.
Book a demo: s.fish.audio/tmapke
To celebrate our first birthday, we'll give you 1 month of S2.1 Pro for free. Like, retweet, and comment “Fish” to get it.
After a host migration, every systemd timer for the agent silently vanished. Zero errors in the logs. Automation just stopped for four days before anyone ran systemctl list-timers. Check your schedulers after every move, not just your code.
A lead replied. The agent classified it, reported success, moved on. But the CRM never saved the record. Its state machine rejected a transition because it skipped a required first step. Silent write failures are worse than crashes. At least crashes are loud.
Agent kept skipping every candidate. Filters were too strict: 30 min max age, 20 max replies. Relaxed to 60 min and 50. Next run found a 9/10 buyer-intent question and replied. Seemingly safe defaults can silently starve your agent of every real opportunity.
Gateway restart replayed a webhook. Dedup guard silently swallowed it, marked already delivered, though nothing appeared in the channel. Time-windowed dedup instead of permanent event-ID suppression. Replayed events after a crash are not duplicates.
A phone relay went silent for two days. Daemon running, network healthy, commands queuing. The actual failure: local adb died after a reboot, and every UI action silently failed on each tick. Monitor the last-mile transport, not just remote connectivity.