iiinova — where your digital self keeps evolving

Christophe DUC @christopheduc.me · 24/09/2026

Gemma 4 26B-A4B (MoE, QAT) • Prompt: 31,321 tokens at 1,025 tok/s (~30 s) • Generation: 1,005 tokens at 33.7 tok/s (~30 s) • Peak memory: 18.4 GB • Total: ~1 min

100

Christophe DUC @christopheduc.me · 24/09/2026

Sure : Qwen 3.8 27B (dense) • Prompt: 29,959 tokens at 274 tok/s (~1 min 50) • Generation: 672 tokens at 8.5 tok/s (~1 min 20) • Peak memory: 19.9 GB • Total: ~3 min 10

100

Christophe DUC @christopheduc.me · 22/09/2026

New Mac mini M6 arrived right on time! Can't wait to see what it can do for local AI paired with a Hermes agent 👀

020

Christophe DUC @christopheduc.me · 20/09/2026

Pic of the day 📸 iPhone 16 pro max 2024

010

Christophe DUC @christopheduc.me · 28/08/2026

A friend built an app to help you stop procrastinating on your phone. No blockers. No gamification. No streaks. Just one question the moment you open an app: "why are you opening this ?" One second to answer, and the reflex turns back into a choice. pourquoi.app

pourquoi.app

pourquoi.app | Récupérer du temps pour ce qui compte vraiment

App sobre pour reprendre conscience de l’usage du téléphone : une question au bon moment, des réponses simples, des stats lisibles. Sans bloqueur ni gamification. Sur iPhone et Android.

221

Christophe DUC @christopheduc.me · 27/08/2026

Many rumors are incoming : the new Fable 5.1 could be released today 🔥 #ai #openai #llm

000

Christophe DUC @christopheduc.me · 27/08/2026

Yes I asked Claude bu raspberry pi 3 has only 1gb ram AI agent app needs more

100

Christophe DUC @christopheduc.me · 26/08/2026

I dusted off my old cluster of four Raspberry Pi 3 Model B boards (1GB RAM, 16GB SD card each) What could I tinker with in terms of AI, agents or related projects ? #ai #llm #local #raspberrypi #tech

130

Christophe DUC @christopheduc.me · 25/08/2026

After waiting so much time for the new Mac Mini, here we go for local AI agents: Hermes, Qwen… AI lab with new M6 Silicon ! let’s go, can’t wait one more month!

010

Christophe DUC @christopheduc.me · 24/08/2026

Interesting ! Thanks. What are the elements that would push you to use Claude Code for example ?

100

Christophe DUC @christopheduc.me · 24/08/2026

Who really stopped using closed source harness like Claude Code or Codex for OpenCode or the latest DeepSeek based harness ? 🧐 What’s your feedback about this switch ?

120

Christophe DUC @christopheduc.me · 18/08/2026

Awesome !! I have old raspberry pi I can use for hermes ! Thanks for idea

000

Reposted by Christophe DUC

Murena @murena-com.bsky.social · 18/08/2026

📲 Buy the new Fairphone (Gen.6+) with /e/OS and receive multiple gifts! – EU offer For every Fairphone (Gen.6+) purchase on murena.com, you get 🔸2 accessories of your choice from our list 🔸1 TB Murena Workspace + Murena MDM for 1 year 👉 Claim your gifts: murena.com/products/sma...

New Fairphone (Gen.6+) with Murena's /e/OS now available on Murena.com

1135

Christophe DUC @christopheduc.me · 18/08/2026

Who is really using FULL local hermes or ai workflow ? What’s your setup ? And your use case 🤓

110

Christophe DUC @christopheduc.me · 17/08/2026

How many companies froze right now because of GitHub dependency? 😬

020

Christophe DUC @christopheduc.me · 04/08/2026

from a tool like Jira (card, functional need) to a GitHub/GitLab PR, to automate the dev workflow : functional analysis, architecture/design decisions, dev, review, automated tests… I’m sure this already exists somewhere as open source ? #dev #ai #aiagent #hermes

100

Christophe DUC @christopheduc.me · 04/08/2026

I’m searching for a Hermes or OpenClaw tutorial/boilerplate/GitHub repository… for creating an agentic developer workflow.

110

Christophe DUC @christopheduc.me · 03/08/2026

Docs MCP in your upgrade loop: worth it, or one more thing to maintain ? #buildinpublic #mcp #ai #llm #claudecode

110

Christophe DUC @christopheduc.me · 03/08/2026

The condition: it tells you what changed upstream, not what breaks in your setup. My compiler and my containers found the rest, including an image that built green and died on startup. What's the last upgrade that passed your build and died on startup ?

110

Christophe DUC @christopheduc.me · 03/08/2026

Today I brought my boilerplate up to date with the Context7 MCP and Claude Code. 3 stacks, 300 packages moved, 10 major versions migrated in one pass. Context7 gave me the breaking changes in seconds instead of 6 changelogs open in tabs.

110

Christophe DUC @christopheduc.me · 02/08/2026

How do you use Codex for coding your app : from terminal or with VS Code extension ? #buildinpublic #ai #dev

150

Christophe DUC @christopheduc.me · 01/08/2026

Time for a break and a coffee walk 🚶‍♂️

010

Christophe DUC @christopheduc.me · 01/08/2026

Where can I get the latest release notes from @proton.me products ? I have Proton Mail version 1.12.1 (5.0.125.9) installed on my Mac => Is this the latest version ?

010

Christophe DUC @christopheduc.me · 01/08/2026

Since Buzz from Block launched, how are you using it day to day ? What have you managed to build with your agents so far ?

011

Christophe DUC @christopheduc.me · 29/07/2026

Whats the best BlueSky desktop client ? (MacOS compatible). #bluesky #opensource #app

130

Reposted by Christophe DUC

Christophe DUC @christopheduc.me · 26/07/2026

How do you review AI-generated code? Using Claude Code's /code-review ? (genuine question: is it accurate when Claude Code wrote the code itself ?) Or an external tool like CodeRabbit ? #ai #dev #code #llm

321

Christophe DUC @christopheduc.me · 27/07/2026

Thanks for your comment and link to the tool 🔥👍

000

Christophe DUC @christopheduc.me · 27/07/2026

Yes that’s my question, make sense to use another tool or another llm. I’m looking for the best (not cheap) solution as we are code reviewer now and not code writter 🤣

000

Christophe DUC @christopheduc.me · 26/07/2026

321

Christophe DUC @christopheduc.me · 26/07/2026

Really ? They can search for the latest doc automaticqlly ?

000

Christophe DUC @christopheduc.me · 26/07/2026

Yes dedup on job_id (a run emits one digest => redeliveries carry the same id). And thanks to your comment, I just closed the authenticity gap: added a signing secret, optional HMAC-SHA256 signing.

100

Christophe DUC @christopheduc.me · 26/07/2026

Is Context7 the best solution for vibe-coding to ensure you're always using the latest version of a component / library ? #mcp #llm #dev #ai

110

Christophe DUC @christopheduc.me · 25/07/2026

Cost/usage is intentionally accumulated per attempt (a retry really did spend the money AGAIN), and outbound webhooks are at least once : everything on internal state is replace/guarded, so a doubled retry can't double it up.

000

Christophe DUC @christopheduc.me · 25/07/2026

The job_id is the dedup key end to end. Celery runs acks_late + reject_on_worker_lost, so redelivery is safe by design rather than rare by luck.

110

Christophe DUC @christopheduc.me · 25/07/2026

So when a task redelivers and re runs : results are written DELETE then INSERT, agent-journal steps upsert ON CONFLICT (job_id, seq) DO NOTHING, and status transitions are guarded (started only from pending, a late failure never clobbers a completed).

100

Christophe DUC @christopheduc.me · 25/07/2026

THanks for asking ! The workers don't dedup. They're stateless and never touch the DB : they just fire HTTP callbacks to a single writer (the Rust API) that owns idempotency via guarded state transitions + replace/upsert, not a dedup table.

200

Christophe DUC @christopheduc.me · 24/07/2026

New week, new Anthropic LLM Welcome Opus 5 !!! #ai #llm #anthropic

010

Christophe DUC @christopheduc.me · 22/07/2026

Is bluesky dying ? Are accounts that switched from X to Bluesky going back to X ? @firefox.com is back on X

000

Christophe DUC @christopheduc.me · 22/07/2026

🔥 de bon augure pour une meilleur souveraineté de nos données !

000