get_started
the data infrastructure for AI Agents in production
Serve your agents with the context they need and observe their traces in production. GlassFlow is one data layer for both sides of the agent. GlassFlow Tares feeds agents correlated data, and GlassFlow Rius traces and debugs every run.
Works with the tools you already use.
Thomas Dohmke
Co-founder & CEO
Entire
products
GlassFlow Tares
Instead of your agent firing dozens of live tool calls at runtime (slow, expensive, inconsistent), GlassFlow Tares prepares and delivers exactly the data each agent needs, already correlated across every source. One clean read replaces many brittle ones: less latency, fewer tokens, more reliable decisions.
Open source, self-hostable, and built on a production streaming core.
github
learn_more
products
GlassFlow Rius
Catch and debug agent failures in production.
Once agents are live, GlassFlow Rius tells you what they did, why they failed, and how quality is changing. Rius is built for any agent, running from short chat sessions to long-running agents that span across hours or days.
Live metrics on running agents (not just the completed minority), heartbeat detection that flags a frozen agent instead of hiding it, and bidirectional MCP so Claude Code, Cursor or ChatGPT Codex can query a failure and open a PR with the fix.
Every trace kept as long as you need it, not capped at 30 days
From install to first result in minutes.
Pick a product and follow the quickstart.
Both are copy-paste and run on your own machine, no waitlist. Up and running in minutes.
glassflow-tares
Deliver your first correlated data package. Install, connect your sources, and serve one clean read to an agent.
>
Cut token spend:
one clean read, not dozens of tool calls
>
Kill latency:
data's ready before the agent asks
>
One connection:all your sources, unified
open the tare’s quickstart
glassflow-rius
See your first agent run appear live.
Drop in the OTEL receiver, run your agent, and watch the trace stream in.
>
Long-running agents:
trace hours-long runs, lose nothing
>
MCP both ways:
agents send traces; coding agents query back
>
Replay any run:
reconstruct what happened last week
open the rius quickstart
Frequently Asked Questions
What is GlassFlow?
GlassFlow is the data infrastructure for AI agents in production. It has two products: GlassFlow Tares, which feeds agents correlated data from every system they touch, and GlassFlow Rius, which traces and debugs what agents do once they’re running.
What’s the difference between GlassFlow Tares and GlassFlow Rius?
GlassFlow Tares handles the data going into the agent, preparing and delivering exactly what each agent needs. GlassFlow Rius handles the data coming out: tracing, metrics, and debugging for agents in production. You can use either on its own or both together.
Is GlassFlow open source?
Yes. GlassFlow Tares is open source and self-hostable on your own infrastructure. A managed Cloud tier is available when you’d rather not run it yourself.
What makes GlassFlow different for long-running agents?
Most observability tools were built for short chat sessions and only compute metrics on completed traces. GlassFlow is built for agents that run for hours or days with live metrics on running agents, heartbeat detection for frozen ones, and unlimited retention so you can always reconstruct what happened.
Ready to give your agents better data and total recall?
Start with Tares
Feed your agents the right data.
get_started
Start with Rius
See and debug agents in production.
get_started
2026 - Copyright GlassFlow.ai
products
contact
docs
get_started
the data infrastructure for AI Agents in production
Serve your agents with the context they need and observe their traces in production. GlassFlow is one data layer for both sides of the agent. GlassFlow Tares feeds agents correlated data, and GlassFlow Rius traces and debugs every run.
Works with the tools you already use.
Thomas Dohmke
Co-founder & CEO
Entire
products
GlassFlow Tares
One correlated read over every system your agents touch.
Instead of your agent firing dozens of live tool calls at runtime (slow, expensive, inconsistent), GlassFlow Tares prepares and delivers exactly the data each agent needs, already correlated across every source. One clean read replaces many brittle ones: less latency, fewer tokens, more reliable decisions.
Open source, self-hostable, and built on a production streaming core.
products
GlassFlow Rius
Catch and debug agent failures in production.
Once agents are live, GlassFlow Rius tells you what they did, why they failed, and how quality is changing. Rius is built for any agent, running from short chat sessions to long-running agents that span across hours or days.
Live metrics on running agents (not just the completed minority), heartbeat detection that flags a frozen agent instead of hiding it, and bidirectional MCP so Claude Code, Cursor or ChatGPT Codex can query a failure and open a PR with the fix.
Every trace kept as long as you need it, not capped at 30 days
From install to first result in minutes.
Pick a product and follow the quickstart.
Both are copy-paste and run on your own machine, no waitlist. Up and running in minutes.
glassflow-tares
Deliver your first correlated data package. Install, connect your sources, and serve one clean read to an agent.
>
Cut token spend:
one clean read, not dozens of tool calls
>
Kill latency:
data's ready before the agent asks
>
One connection:all your sources, unified
open the tare’s quickstart
glassflow-rius
See your first agent run appear live.
Drop in the OTEL receiver, run your agent, and watch the trace stream in.
>
Long-running agents:
trace hours-long runs, lose nothing
>
MCP both ways:
agents send traces; coding agents query back
>
Replay any run:
reconstruct what happened last week
open the rius quickstart
Frequently Asked Questions
What is GlassFlow?
GlassFlow is the data infrastructure for AI agents in production. It has two products: GlassFlow Tares, which feeds agents correlated data from every system they touch, and GlassFlow Rius, which traces and debugs what agents do once they’re running.
What’s the difference between GlassFlow Tares and GlassFlow Rius?
GlassFlow Tares handles the data going into the agent, preparing and delivering exactly what each agent needs. GlassFlow Rius handles the data coming out: tracing, metrics, and debugging for agents in production. You can use either on its own or both together.
Is GlassFlow open source?
Yes. GlassFlow Tares is open source and self-hostable on your own infrastructure. A managed Cloud tier is available when you’d rather not run it yourself.
What makes GlassFlow different for long-running agents?
Most observability tools were built for short chat sessions and only compute metrics on completed traces. GlassFlow is built for agents that run for hours or days with live metrics on running agents, heartbeat detection for frozen ones, and unlimited retention so you can always reconstruct what happened.
Ready to give your agents
better data and total recall?
Start with Tares
Feed your agents the right data.
get_started
Start with Rius
See and debug agents in production.
get_started
2026 - Copyright GlassFlow.ai
products
contact
docs
get_started
the data infrastructure for AI Agents in production
GlassFlow is one data layer for both sides of the agent: GlassFlow Tares feeds agents correlated data, and GlassFlow Rius traces and debugs every run.
Works with the tools you already use.
Thomas Dohmke
Co-founder & CEO
Entire
products
GlassFlow Tares
One correlated read over every system your agents touch.
Instead of your agent firing dozens of live tool calls at runtime (slow, expensive, inconsistent), GlassFlow Tares prepares and delivers exactly the data each agent needs, already correlated across every source. One clean read replaces many brittle ones: less latency, fewer tokens, more reliable decisions.
Open source, self-hostable, and built on a production streaming core.
products
GlassFlow Rius
Catch and debug agent failures in production.
Once agents are live, GlassFlow Rius tells you what they did, why they failed, and how quality is changing. Rius is built for any agent, running from short chat sessions to long-running agents that span across hours or days.
Live metrics on running agents (not just the completed minority), heartbeat detection that flags a frozen agent instead of hiding it, and bidirectional MCP so Claude Code, Cursor or ChatGPT Codex can query a failure and open a PR with the fix.
Every trace kept as long as you need it, not capped at 30 days
From install to first result in minutes.
Pick a product and follow the quickstart.
Both are copy-paste and run on your own machine, no waitlist. Up and running in minutes.
glassflow-tares
Deliver your first correlated data package. Install, connect your sources, and serve one clean read to an agent.
>
Cut token spend:
one clean read, not dozens of tool calls
>
Kill latency:
data's ready before the agent asks
>
One connection:all your sources, unified
open the tare’s quickstart
glassflow-rius
See your first agent run appear live.
Drop in the OTEL receiver, run your agent, and watch the trace stream in.
>
Long-running agents:
trace hours-long runs, lose nothing
>
MCP both ways:
agents send traces; coding agents query back
>
Replay any run:
reconstruct what happened last week
open the rius quickstart
Frequently Asked Questions
What is GlassFlow?
GlassFlow is the data infrastructure for AI agents in production. It has two products: GlassFlow Tares, which feeds agents correlated data from every system they touch, and GlassFlow Rius, which traces and debugs what agents do once they’re running.
What’s the difference between GlassFlow Tares and GlassFlow Rius?
GlassFlow Tares handles the data going into the agent, preparing and delivering exactly what each agent needs. GlassFlow Rius handles the data coming out: tracing, metrics, and debugging for agents in production. You can use either on its own or both together.
Is GlassFlow open source?
Yes. GlassFlow Tares is open source and self-hostable on your own infrastructure. A managed Cloud tier is available when you’d rather not run it yourself.
What makes GlassFlow different for long-running agents?
Most observability tools were built for short chat sessions and only compute metrics on completed traces. GlassFlow is built for agents that run for hours or days with live metrics on running agents, heartbeat detection for frozen ones, and unlimited retention so you can always reconstruct what happened.
Ready to give your agents better data and total recall?
Start with Tares
Feed your agents the right data.
get_started
Start with Rius
See and debug agents in production.
get_started
2026 - Copyright GlassFlow.ai