Orchestrating the next era of AI agents ⚙️ | @ApacheAirflow Committer & PMC member | Sr Eng Director @ @astronomerio | Open-source advocate
London
Joined June 2009
- Tweets1.8K
- Following2.2K
- Followers1.6K
- Likes522
Kaxil Naik retweeted
PHP sites behind Cloudflare already have an initial set of WAF rules deployed for the known exploit vectors in these vulnerabilities. The rules are available to all customers, including free.
We are seeing early attacks emerging and are updating mitigations as new vectors show up.
This is a STOPGAP only. Patch as soon as @ThePHPF releases the update at 14:00 UTC today.
📣📢 Airflow AI Provider 1.0 is out.
AI agents as @ApacheAirflow tasks, with your warehouses, buckets and APIs as their tools. The model never sees a credential.
Any agent framework, approval before it acts, smarter retries, and it runs on Airflow 2.11+ 🧵
Retry policies work with Jev-like Decision models on any task, AI or not: a model reads the error and decides whether to retry or fail now.
A revoked key failed at once instead of burning 3 retries. A 429 retried 60s later and passed.
Full post, with every Dag we ran: airflow.apache.org/blog/comm…
38 people built this between April and October. Thank you all 🙏
Kaxil Naik retweeted
What if Spotlight actually understood the meaning inside your files?
DigUp is an open-source macOS menu bar app powered by EmbeddingGemma 2 that brings 100% local, multimodal semantic search to your Mac.
Instead of searching for file names, you can use natural language to:
📸 Find specific scenes in your photos
⏳ Jump to the exact timestamp in audio or video
💻 Search through your codebase by intent
Everything runs locally in a single vector space.
Loops 🔁 ..
Will be part of @ApacheAirflow 3.4, planned to release in next couple of weeks.
Six open Decision Models shipped in the 16 days after @typesafeai 's Jev: Laya, decider, Kev, GLiNER2.5-Decide, Strands Decider & Clef.
We ran Jev and seven open models on one Airflow job, deciding whether a failed task should retry or stop, using 400 errors from GitHub issues.
One threshold for every model doesn't work. Median confidence was 0.94 for Jev and 0.45 for GLiNER, so 0.8 means something different to each. We let every model make the same number of stops and counted the wrong ones.
Write-up, including how we built the test set: astronomer.io/blog/how-to-ev…
Kaxil Naik retweeted
Mods are absolutely insane. You can now customize Claude to work and look the way you want by just prompting it.
Each person works differently, so there's no reason why everyone should have an identical Claude experience. Make Claude your own, and share mods as plugins so others can try your mods too.
Kaxil Naik retweeted
Introducing Gemini 4 Argon – our new frontier model.
It’s built for complex workflows across coding, enterprise knowledge work, and cybersecurity defense – rolling out today to a set of trusted testers through our Fairwind Program.
Kaxil Naik retweeted
wrote up some thoughts about jev & data engineering after playing with it this weekend - feels like it's going to be big
Kaxil Naik retweeted
Introducing Jev Model Router for Claude Code
This Claude Code Mod lets you use Jev through its direct @typesafeai API or @vercel AI Gateway
With every request you send to Claude Code, Jev classifies the subagent model, main model (only at session start to avoid breaking the cache), and effort level
Install it with one command:
npx claude-code-templates@latest --mod productivity/jev-model-router
Full Mod details:
aitmpl.com/component/mod/pro…
Kaxil Naik retweeted
Show Codex a workflow once. Reuse it as a skill.
Record & Replay lets you show Codex a recurring task, like filing an expense report or submitting a time-off request.
Codex turns that demo into an inspectable, editable skill.
You control when recording starts and stops.
Hard-coding MCP server URLs and tokens in DAG files is how credentials end up in git.
MCPToolset in the Common AI @ApacheAirflow provider connects agent tasks to any MCP server through something Airflow already does well: Connections.
How it works:
🔌 Point an AgentOperator task at an MCP server by connection ID. The URL, transport, and auth token live in the connection or your secrets backend.
🔀 Streamable HTTP, SSE, and stdio transports are all supported. Switch by editing the connection; no DAG change.
🧩 Need several servers in one task? Add a tool_prefix to each so tool names never collide.
🔐 For endpoints that want short-lived auth (OAuth, signed JWTs, Workload Identity), pass a token_provider callable. A fresh bearer token is minted on every connection and masked in task
logs automatically.
The same dev-to-prod promotion story you have for databases now applies to agent tooling: swap the connection, ship the same DAG.
Kaxil Naik retweeted
Welp, that happened faster than I predicted. Thought it would be end of 2027, then early 2027, but agentic traffic growing so fast that bots have now passed human traffic online for the first time in the Internet's history. radar.cloudflare.com/traffic…
