GlassFlow

GlassFlow builds the data infra for AI agents in production. Tares feeds agents correlated data from every system & Rius traces and debugs what prod agents do.

[GlassFlow @glassflowdev]

For the full write-up on the 7 failure modes of AI Agents see here: https://t.co/k2YNo0AmTe

[GlassFlow @glassflowdev]

6 of the 7 ways #AIAgents fail in production raise no error.
It doesn't crash—it finishes, confidently wrong.
The 7:
• Tool misuse
• Context loss
• Goal drift
• Retry loops
• Frozen agents
• Cascading failures
• Silent degradation

[GlassFlow @glassflowdev]

How do you build a #datastore for agents, not analysts? 🦆
This Thurs (Sep 17), we show an open-source, one-file #DuckDB store where the agent gets woken up with context attached and writes its findings to the same table.
Last few spots: https://t.co/yzjIQIei3q
#AIAgents #agentic

[GlassFlow @glassflowdev]

Grab the community offer for 3 months free Rius 👇
https://t.co/Ryd837t5iY

[GlassFlow @glassflowdev]

We were at the AI Builders Monthly event in London last night 🚀
The demo: an AI agent sends its traces to Rius, finds why it's failing, and ships its own fix to GitHub.

Thanks for a great night to everyone who attended!

[GlassFlow @glassflowdev]

Agents fail in production. #Rius helps them recover automatically. 🚀

Live demo tonight at AI Builders Monthly London.
👉 https://t.co/YoCuR4hyhf

#AIAgents #AIBuilders #agentobservability

[GlassFlow @glassflowdev]

"How AI Agents Fix Their Own Production Failures"

AI Builders Monthly London, Sept 10: Our founder, Armend will show how agents use #observability to find the root cause of a prod failure and fix it at runtime.
Plus a first look at Rius 👀
🎟️ https://t.co/lufB621u9h
#AIAgents

[GlassFlow @glassflowdev]

Thanks to all who showed up for Pipes, Streams, and Queries with @ClickHouseDB last week 🙌

Sasi Teja broke down how to turn raw #OpenTelemetry streams into real-time insights in #ClickHouse.

Want to try it yourself?
Full demo here 👇 https://t.co/XJLXxF7lYN

[GlassFlow @glassflowdev]

Less than 10 days left! 🕒
Join us for our upcoming meetup with @ClickHouseDB and @platformatory!

Don't miss Sasi's session on turning raw telemetry streams into actionable insights at scale using #OpenTelemetry & #ClickHouse.

📅 July 11
🎟️ Register: https://t.co/JbXdcGkWar

[GlassFlow @glassflowdev]

How do enterprises handle #ClickHouse ingestion at TB scale?
– Pre-ingestion dedup (no FINAL queries)
– Schema evo that doesn't break CDC
– Backfill + live sync running simultaneously
– Linear scaling: 500K+ rps, 44TB/day

The pattern that's working → https://t.co/xRdpRonUna

[GlassFlow @glassflowdev]

We're speaking at Pipes, Streams & Queries on July 11 in Bengaluru! 🇮🇳
Joining forces with @ClickHouseDB & Bangalore Streams for a morning of real-world talks on data ingestion & fast analytics at scale.
Local to Bengaluru?
Come hang 👇
https://t.co/JbXdcGltZZ

#ClickHouse

[GlassFlow @glassflowdev]

Running #ClickHouse in production?
We documented the 10 mistakes that break #datapipelines incl. #deduplication, sharding traps, disaster recovery gaps, and more.

Free guide, with fixes 👇

https://t.co/SRHTrndeDK

#realtimedata

[GlassFlow @glassflowdev]

Full tutorial — OTel → GlassFlow → ClickHouse pipeline, Kubernetes + Helm setup, real dataset included: https://t.co/i2CoFP1oCf

[GlassFlow @glassflowdev]

Your OTel Collectors are double-counting your 5G RAN metrics. Here's the proof — and how to fix it 🧵

[GlassFlow @glassflowdev]

The result: surviving_dupes = 0. And you get to use a simple MergeTree in ClickHouse instead of fighting ReplacingMergeTree.

[GlassFlow @glassflowdev]

ClickHouse mistake #5: writing JOINs like it's #PostgreSQL.
Rules:
→ Smaller table always on the RIGHT (it loads into memory)
→ Denormalize upstream instead of joining at query time
→ Use Dictionaries for slow-changing dimensions, not JOINs
For more: https://t.co/cWU1KJH9Ud

[GlassFlow @glassflowdev]

This is one of 7 ingestion problems that break most ClickHouse projects in the first 30 days.
We added all of them (and how to solve them) here: https://t.co/kwDdCXPIKI

[GlassFlow @glassflowdev]

Running a historical backfill into #ClickHouse while #CDC is live?
There's a race condition that will corrupt your data, and it's easy to miss until your dashboards are wrong.🧵

[GlassFlow @glassflowdev]

The right fix: track timestamps and primary keys before writing to #ClickHouse.
Only write a record if it's newer than what's already there. Regardless of which pipeline sent it.
This keeps both pipelines running safely in parallel.