I just open-sourced Agentura — Kubernetes for AI agent skills.
Skills are Markdown. Agents run in ephemeral K8s pods. Corrections compound into permanent memory. Self-hosted.
Here's what it does 🧵
Trigger from anywhere:
→ Slack (@mention or /agentura)
→ GitHub webhooks (PR opened → review pipeline)
→ Cron schedules
→ REST API
→ CLI
One skill. Five entry points. All config.
The real differentiator: memory that compounds.
Correct an agent once →
→ Reflexion rule generated
→ Injected into every future run
→ Regression test auto-created
Same mistake never happens twice.
Six months in, your agents know your domain better than any SaaS could.
Every AI agent framework wants you to write Python classes.
Agentura: write a https://t.co/8f6AbkUhJZ + a YAML config. That's it.
Pick your executor:
→ PTC Worker (200MB) — lightweight tool-calling
→ Claude Code Worker (800MB) — full sandbox, bash, file I/O
→ Legacy — in-process, no pod needed
No boilerplate. No framework code.
@policybazaar This is not an isolated issue — it’s a pattern of malpractice where reimbursement cases are brushed aside casually while marketing “high claim ratios.”
Requesting @irdai to audit such practices & protect policyholders from systemic claim denial. #InsuranceReform
@Niva_Bupa@policybazaar@NivaBupaSupport deeply disappointed with how my genuine claim (ID: 25565354) was rejected despite clear medical justification from treating doctor. CRP was 90 (normal 6), patient was dehydrated with bronchitis IV fluids & nebulisation were essential.
@policybazaar Escalating to @irdai@jagograhakjago — @NivaBupa agents are dismissing genuine claims with shocking insensitivity. Case ID: 25565354.
Told “CRP 90 is nothing” & “we don’t care if anything happens” — zero empathy, zero accountability. Patients deserve better. #ConsumerRights
@Niva_Bupa It’s concerning that companies promoting “high claim settlement ratios” reject medically valid emergencies on technicalities.
This erodes public trust and raises questions about how claim data is presented to customers. Hoping for a transparent review. #ConsumerRights
@pharmeasyapp doing shady things of delivering lesser quantities of ordered medicines and denying proof . Never expected this from @pharmeasyapp highly unprofessional and disgraceful
A few months ago, we embarked on a journey to have a 10x improvement in the DevEx for our engineers. We're proud to say that we overshot our own expectations.
Here's @srinidhis05 and @eigenmind discussing this at @KubeCon_
https://t.co/1UUPyBuwib
Great insight into development challenges at @Razorpay scale! Using Kubernetes and cloud is good, but there is still friction to be overcome that can waste developer time
Venkat and Srinidhi at #KubeCon
I need to bookmark @Razorpay team's excellent #KubeCon presentation for a re-watch.
They tackled these problems:
1: Bypass CI/CD loop
2: Network limitations
3: Avoid container restarts
4: Service dependencies
5: Provision ephemeral infra
6: Route traffic to the right service
All of our work is open-source: https://t.co/ceovq1PVd4
Join our Platform engineers @srinidhis05 and @eigenmind at #KubeCon 2021
-- they'll be speaking about the entire stack and answering questions.
https://t.co/TuzRJx0rtO
Our engineers struggled with unstable and sparsely available stage environments, along with high build & deploy times. Here’s how we built a @kubernetesio based workflow that allowed them to iteratively develop cloud apps on their laptops.