We're about to ship a ground-up redesign of Capture workflows that makes them fast to build, easy to read, and ready for AI agents to create and manage programmatically.
bitdrift Capture workflows have always been powerful. We're about to ship a ground-up redesign that makes them fast to build, easy to read, and (when we release our public API in the very near future) ready for AI agents to create and manage programmatically. This redesign lays the groundwork for first-class tooling support: a public API, CLI access, and the ability for LLM agents to instrument apps and manage workflows autonomously.
A quick refresher
Capture is a fundamentally different take on mobile observability. Rather than sending non-useful samples or firehosing every log from every device to a backend that processes and discards most of it, Capture keeps data on-device and only ships it when something interesting happens. The mechanism that defines "interesting" is the workflow: a state machine deployed in real time to your fleet that describes sequences of events to match on and actions to take when they do.Workflows are powerful. They can also be humbling. Over time, as customers pushed them to their limits, we heard the same feedback again and again: too many clicks, too many advanced features, too much nuance to learn before being productive. We took that feedback seriously and rebuilt the workflow experience from the ground up.
What's coming
A faster, cleaner builder
The new builder is optimized for speed. Graph layout is now calculated automatically, drag and drop is gone, and steps are placed contextually as you build. Keyboard shortcuts cover the most common operations, so power users can move quickly without reaching for the mouse. The result is an experience that feels much more like writing code than assembling a flowchart.
Actions are attached, not floating
In the previous builder, actions (record session, plot chart, etc.) were standalone floating nodes connected to matchers by edges. This made complex workflows hard to read at a glance. Actions are now directly attached to the step that triggers them, so the relationship between a match condition and its outcome is immediately obvious.
Steps have names. Everything follows.
You can now name your workflow steps, and all titles, series names, and chart metadata are inferred from those names automatically. Less busywork, and your charts finally describe what they're actually measuring.
Exit conditions
We're generalizing and replacing the previous "timeout rule" with a more powerful and flexible concept: exit conditions. An exit condition can be a timeout or any matcher, and when it fires, it can trigger any combination of actions (charts, record sessions, etc.) just like a normal match step. This makes it straightforward to express patterns like "if the user abandons the checkout flow at any point, record the session."
Reset and entrance behavior, made visible
Reset and entrance behavior — the rules that govern when a workflow restarts or re-enters — are now surfaced clearly in the UI rather than buried in configuration drawers. Workflows can now contain multiple independent flows, each with its own entry point, running in parallel. Resets on one flow are isolated from all others.
Smarter conditionals
Match conditions now support both ANDs of ORs and ORs of ANDs. Previously, only one form was supported, which forced awkward workarounds for common cases. Both are now first-class.
Charts, simplified and split out
The previous "chart" action covered a lot of ground in a single configuration surface. We've split it into focused types (Counters, Histograms, Rate, Average, Funnel, and Sankey), each with a vastly simpler configuration. The "Advanced charts" section is gone. Chart customization (titles, series names, units) is now done directly on the chart itself rather than in the workflow, which means the same underlying data can be visualized differently depending on where it's shown.
Measure duration gets more powerful
The measure duration step now supports restricting actions to a specific duration range. For example, only record a session when the time between two steps exceeds 10 seconds. You can also attach any existing action to a measure duration step, giving you full flexibility over what happens when a timing threshold is crossed.
The big reason we rebuilt this
Customer feedback drove a lot of this. We heard clearly that workflows needed to be more intuitive, faster to learn, and less prone to surprising behavior. Every item in the list above traces back to a real conversation with a real user who got stuck.But there's a bigger reason.We're about to launch a public API and CLI that unlock everything the UI can do (and more). AI agents will be able to instrument apps, create workflows, and debug issues autonomously. For that to work, workflows had to be representable as clean, predictable code rather than a sprawling visual graph with implicit rules and hidden state. The redesign wasn't just about making workflows easier for humans. It was about making them machine-readable.This is the foundation for what comes next.
