WeaveScope

WeaveScope is an observability platform built specifically for AI agents running on Elixir and the BEAM. If your agents live in the Elixir ecosystem, WeaveScope gives you tracing, debugging, and monitoring that understands how those systems actually behave, instead of forcing you to squint at a generic APM dashboard or an LLM observability tool that assumes everyone writes Python.

The gap it fills is real. The mainstream AI observability stack has grown up around Python and TypeScript, so teams building agents on the BEAM have been left stitching together telemetry by hand or going without. That's a bad place to be with agent workloads, where a single run can fan out into dozens of model calls and tool calls, costs accumulate token by token, and the interesting failures happen three steps deep in a chain you can't see. WeaveScope treats the Elixir agent as a first-class citizen and instruments that whole chain.

Integration goes through a single dependency called BeamWeaver, published on Hex. You add it to your mix project, configure an API key and endpoint, and it starts exporting instrumented agent runs, model calls, and tool calls in the background. There's no separate collector to deploy and no manual span plumbing to write, the library handles the export asynchronously so it stays out of the way of your actual agent code. Documentation lives at docs.weavescope.com for the configuration details beyond that one-line pitch.

Once traces are flowing, WeaveScope reconstructs each run as a unified timeline, so you can inspect every model call, tool call, and span in the order they happened and see exactly where a run went sideways. That's the debugging half, and it's the part that changes day-to-day work the most. Instead of grepping logs and mentally reassembling what an agent did across a dozen processes, you get the whole run laid out as one coherent story. The monitoring half tracks throughput, latency, errors, token usage, and costs across your agents over time, which turns questions like which agent is burning the budget or where latency crept in from guesswork into a dashboard lookup. Token and cost tracking in particular is something generic APM tools simply don't have a concept for, and it's often the metric that decides whether an agent feature is economically viable at all.

It's aimed at development teams building AI agents on Elixir and the BEAM who need visibility into what their agents are doing in production. That includes teams debugging misbehaving runs, teams optimizing cost per task, and anyone who wants error and latency trends before users start reporting them. The Elixir AI ecosystem has been growing steadily, and the teams betting on it tend to be exactly the kind of pragmatic engineering shops that expect real observability from day one rather than treating it as a later problem. Alerts are listed as coming soon, building on the tracing and monitoring data that's already collected, so proactive notification is on the roadmap rather than in the product today. If alerting is a hard requirement right now, that's worth knowing before you commit, though the underlying data it will draw on is already being captured.

What makes it different is the focus. There are plenty of LLM observability platforms, but almost none of them speak Elixir natively, and generic BEAM monitoring tools don't understand agents, model calls, or token costs. WeaveScope sits precisely at that intersection, and the one-line integration reflects it. Because the instrumentation library is built for the runtime you're already on, you're not writing adapter glue or shipping traces through a sidecar meant for another ecosystem. For a niche this specific, being the tool that actually fits the stack is the whole pitch, and it's a credible one given how much friction the alternative involves. Anyone who has tried to bolt a Python-first tracing SDK onto a BEAM application knows the pain being avoided here.

Pricing follows a straightforward tiered model. The free plan costs nothing and covers 100K events per month with 14 days of retention and two users, which is enough to instrument a real project and live with the product for a while. Pro runs 29 euros per month for 1M events, 30 days of retention, and unlimited users, Scale is 99 euros for 10M events and 90 days of retention, and an Enterprise tier handles 50M+ events with custom retention and terms through a talk-to-sales motion. Event-based pricing maps cleanly onto agent workloads, since busy agents generate more spans, and the free tier means an Elixir team can wire in BeamWeaver this afternoon and see their first traced run before deciding whether to pay anything at all.