Show HN: WeaveScope – Elixir native observability for AI agents

caudena1 pts0 comments

Skip to content

WeaveScope is in early access — tracing and monitoring for Elixir agents

Find the failing step.

Trace the model or tool call behind failed, slow, or expensive Elixir agent runs.

Start free

Book a production trace review

Free up to 100K accepted events / month · no credit card · one Elixir dependency

Tracing

Inspect model calls, tool calls, and spans across instrumented agent runs. How it works →

Filters

Last 30 days

All<br>Errors<br>Slow > 30s

Filter by trace ID, user, model...<br>25 of 25<br>on page<br>Reset

ObservationsNo prior

43,040<br>Analytics unit

Error rateNo prior

0.00 %<br>Traces with errors

p95 latencyNo prior

29.56 s<br>Trace refs

TokensNo prior

5,093,888<br>Input + output

CostNo prior

$11.046970<br>Trace refs

Recent agent traces. Select a row for a detail preview.

Trace ID<br>Name<br>User<br>Model<br>Spans<br>Tags<br>Tokens<br>Cost<br>Total time<br>When

019fd6f2...0ea0x_signal_desk.post_summaryanonymouszai:glm-5.295,646$0.012621.79 s2026-08-06 12:01

019fd6f2...9bcdx_signal_desk.post_summaryanonymouszai:glm-5.295,426$0.012017.71 s2026-08-06 12:01

019fd6d7...60c2x_signal_desk.post_summaryanonymouszai:glm-5.294,365$0.009113.61 s2026-08-06 11:31

019fd684...4fabx_signal_desk.post_summaryanonymouszai:glm-5.297,029$0.017332.78 s2026-08-06 10:01

019fd684...fe35x_signal_desk.post_summaryanonymouszai:glm-5.294,528$0.009318.17 s2026-08-06 10:01

019fd64d...54d5x_signal_desk.post_summaryanonymouszai:glm-5.297,117$0.017940.66 s2026-08-06 09:01

019fd64d...93d5x_signal_desk.post_summaryanonymouszai:glm-5.294,601$0.010023.68 s2026-08-06 09:01

019fd632...9e98x_signal_desk.post_summaryanonymouszai:glm-5.295,593$0.012030.45 s2026-08-06 08:31

Turns

Tree<br>Waterfall

Search observations

⌁x_signal_desk.post_summary<br>21.79 s

⌁PatchToolCallsMiddleware.wrap_model_call<br>21.79 s

⌁DynamicPromptMiddleware.wrap_model_call<br>21.79 s

⌁StructuredOutputRetryMiddleware.wrap_model_call<br>21.79 s

⌁ModelRetryMiddleware.wrap_model_call<br>21.79 s

⌁RuntimeZaiModelMiddleware.wrap_model_call<br>21.79 s

⌁PostSummaryValidationMiddleware.wrap_model_call<br>21.79 s

⌁RunnableSequence<br>21.79 s

•zai:glm-5.2glm-5.2<br>▤ 5.6k21.77 s$0.012589

generation<br>zai:glm-5.2

019fd6f2...1825

21.77 s5,646 tokens$0.012589<br>zai:glm-5.2<br>none

Messages<br>Tools<br>Details

Tools 0 called/0 available

No tool calls were recorded for this observation tree.

Input 2 blocks

System instructionsTextMarkdownXMLJSON

User 019fd6f2...cabdTextMarkdownXMLJSON

Reasoning 1 blocks

Reasoning TextMarkdownXMLJSON

Output 1 blocks

AI 20260806...44d0TextMarkdownXMLJSON

"posts": [<br>"id": 11301,<br>"key_facts": [<br>"Uniswap launched TradePools, a new launchpad on Robinhood Chain",<br>"Platform accessible at pools.trade",<br>"Commentator notes over 170 launchpads already exist, raising scam token concerns"

Tools 0 called/0 available

No tool calls were recorded for this observation tree.

Observation details

Modelzai:glm-5.2<br>Tokens5,646<br>Cost$0.012589<br>Duration21.77 s

Built for how agents actually run

Know what happened.<br>Then know what to fix.

Tracing

Model and tool calls in one timeline

BeamWeaver connects instrumented model calls, tool calls, and spans across an agent run. Jump from a failed step to its inputs, outputs, latency, and cost.

Monitoring

The operational signals that matter

Follow throughput, latency, errors, token usage, and cost as your agents run. Give your team a shared view of the work behind every outcome.

Alerts Coming soon

Get a signal before an agent failure spreads

Alerting is next. We are building it on top of the tracing and monitoring data your team already has—not as a separate black box.

One dependency. Trace context.

Instrument an agent<br>before lunch.

Add the dependency and configure one endpoint. BeamWeaver exports instrumented agent runs, model calls, and tool calls in the background.

View BeamWeaver on Hex

mix.exs<br>runtime.exs<br>agent.ex

defp deps do<br>{:beam_weaver, "~> 0.1.15"}<br>end

config :beam_weaver,<br>openai: [api_key: System.fetch_env!("OPENAI_API_KEY")],<br>weave_scope: [<br>api_key: System.fetch_env!("WEAVESCOPE_API_KEY"),<br>endpoint: "https://app.weavescope.com"

defmodule Support.Agent do<br>alias BeamWeaver.Agent<br>alias BeamWeaver.Core.Message

def run(query, context \\ %{}) do<br>model =<br>BeamWeaver.Models.init_chat_model!("openai:gpt-5.4-mini",<br>temperature: 0.2,<br>timeout: 30_000

{:ok, agent} =<br>Agent.build(<br>name: "support.reply",<br>model: model,<br>system_prompt: "Answer support questions clearly."

Agent.invoke(agent, %{messages: [Message.user(query)]},<br>trace: [<br>name: "support.reply",<br>user_id: context[:user_id],<br>execution_mode: "support_reply",<br>fields: %{ticket_id: context[:ticket_id], account_id: context[:account_id]}<br>end<br>end

Pricing

Simple, predictable.

Start free. Usage counts accepted observation events; a trace can contain multiple events.

Free

Community support

€0/ month

Accepted events100,000

Data/hour100 MB/hour

Retention14 days

Projects1

Users2

OverageNot available

Start free

Pro

In-app support

€29/ month

Accepted events1,000,000

Data/hour2.5...

agent model trace calls post_summaryanonymouszai s2026

Related Articles