Agnost has a model read your conversations. We compute the numbers.
Closest category to ours, opposite method. Agnost analyzes production conversations with a model to find where users get stuck and suggests fixes, and it is genuinely good at that job. smolanalytics is one MIT Go binary that does product analytics and agent observability from the same events, every number computed deterministically, nothing labelled by a model.
smolanalytics vs Agnost, what is the difference?
Both live in agent observability, but they answer differently. Agnost (agnost.ai) pitches itself as “Catch agent failures your evals miss.”: it continuously analyzes your production conversations, extracts intent and sentiment signals, generates failure categories like broken workflows, repeated retries, setup friction and churn risk, and turns the highest-impact patterns into reviewed fixes for your agent. It is OpenTelemetry native, advertises a 2-minute setup with any LLM and any framework, and grew out of MCP-server analytics, its GitHub org bio still reads “Usage Analytics for your MCP Servers”. It is a hosted platform: as published on agnost.ai in July 2026, Free covers 1,000 events/mo with 7-day retention, Starter is $49/mo for 10,000 events with 30-day retention, Pro is $499/mo for 1,000,000 events with 90-day retention, and self-hosting is an Enterprise conversation. smolanalytics takes the opposite method. It is a single MIT Go binary that does web and product analytics (funnels, retention, paths, cohorts, breakdowns, heatmaps, surveys, feature flags and A/B) plus agent observability off the same event stream: tool-call volume, error rate, latency p50/p90/p99, client split, an error taxonomy, and conversation health like turns, re-ask rate, abandonment and resolution. Every one of those numbers is computed from your events, never labelled by a model, and a CI test pins /v1, MCP and the dashboard to the same answer. You ask it questions in plain English from your editor over MCP, 79 tools, one org token for every project, answered by your own coding agent on your own model, so the AI is never metered by us. Pricing is a 14-day full trial with no card, Pro $49/mo for 1M events, Scale $149/mo for 10M, $8 per extra million, self-host free forever.
what each one actually is
“Catch agent failures your evals miss.” Agnost “continuously analyzes production conversations, finds where users get stuck, frustrated, or fail to convert, and turns the highest-impact patterns into reviewed fixes for your agent.”
agnost.ai, homepage hero, Jul 2026
“Works with any LLM, any framework. 2-min setup. OpenTelemetry native.” Its roots are MCP analytics: the GitHub org bio reads “Usage Analytics for your MCP Servers”, wrapping a server with one track() call.
agnost.ai and github.com/AgnostAI
smolanalytics is one MIT Go binary and one data file. Product analytics and agent observability come out of the same events, and every number is computed, no model reads your conversations.
what we do instead
Agnost vs smolanalytics
| Agnost | smolanalytics | |
|---|---|---|
| The job it is built for | conversation insight + suggested fixes | product analytics + agent health |
| How agent numbers are made | a model reads your conversations | computed from events, never model-labelled |
| Intent, sentiment, fix suggestions | core product | not offered, deliberately |
| Failure surface | model-generated failure categories | computed error taxonomy + re-ask, abandonment, resolution |
| Funnels, retention, paths, cohorts | not its category | yes, same binary, same events |
| Instrumentation | OTel native, wrap MCP server with track() | send events, nothing in your request path |
| Self-host | Enterprise tier, custom | one MIT binary, free forever |
| Data retention | 7 to 90 days by plan | your disk, your call |
| 1M events/mo | $499/mo (Pro, published Jul 2026) | $49/mo (Pro) |
| Dashboard vs API drift | hosted dashboard at app.agnost.ai | CI pins /v1 == MCP == dashboard |
when Agnost is the right tool
- You want a model to read your conversations and tell you why users got stuck. Intent and sentiment extraction, generated failure categories like broken workflows and churn risk, and reviewed fix suggestions are Agnost's whole product. We do none of that on purpose: our numbers are computed, so we can promise they never drift, but they will never explain a conversation the way a model reading it can.
- You are instrumenting an MCP server and want the purpose-built view. Agnost started as MCP-server analytics, wraps your server with a single track() call, and third-party writeups (WorkOS, Jul 2026 era) describe tool invocations, latency percentiles, error captures, user journeys and a client-distribution view across Claude Desktop, VS Code and other AI clients. That is a polished, specific lens we respect.
- You want a free tier before paying anything. Agnost's published Free plan covers 1,000 events/mo with 7-day retention. We do not have a free cloud tier, our free path is self-hosting the MIT binary, so if a hosted $0 start is the requirement, that is them. Check agnost.ai for current plans rather than trusting a competitor's snapshot.
Computed, on your own box, in 30 seconds
# no hosted platform, no retention window, no model reading your data docker run -p 8080:8080 -v $PWD/data:/data \ -e SMOLANALYTICS_DB=/data/smolanalytics.data \ ghcr.io/arjun0606/smolanalytics # dashboard + MCP at http://localhost:8080 # then, in your editor: "re-ask rate and tool error taxonomy for the last 14 days, by client"
If you want a model's read on your conversations, Agnost does that well. If you want product analytics and agent health as numbers you can pin in CI, on hardware you own, that is us.
questions
Is smolanalytics an Agnost alternative?
For agent observability, yes, with an honest caveat about method. Agnost analyzes your production conversations with a model: intent and sentiment signals, generated failure categories, and suggested fixes for your agent. smolanalytics computes agent observability from the events your app sends: tool-call volume, error rate, latency p50/p90/p99, client split, an error taxonomy, and conversation health like turns, re-ask rate, abandonment and resolution, and it does full product analytics (funnels, retention, paths, cohorts, heatmaps, surveys, flags and A/B) from the same events. If you want a model's qualitative read and fix suggestions, that is Agnost. If you want deterministic numbers plus product analytics in one MIT binary, that is us.
How is smolanalytics' error taxonomy different from Agnost's failure categories?
Ours is computed, theirs is model-generated. Agnost's homepage says it “reads real production conversations and generates the failure categories that matter to your product: broken workflows, repeated retries, setup friction, churn risk, and more.” That is a model's interpretation, which can be genuinely insightful and can also shift when the model does. Our taxonomy, re-ask rate, abandonment and resolution are derived deterministically from your event stream, nothing is an opinion, and a CI test asserts the /v1 API, the MCP answer and the dashboard return the same number for the same question.
Which costs less at a million events a month?
As published on agnost.ai in July 2026, Agnost's Pro plan is $499/mo for up to 1,000,000 events with 90-day retention, and its Starter is $49/mo for up to 10,000 events. smolanalytics Pro is $49/mo for 1M events with no retention window imposed by us. Prices change, so verify theirs on agnost.ai before deciding, but that is the published gap today, and part of why: their platform runs model analysis over your data, ours computes, and computing is cheap.
Can I self-host Agnost? Is it open source?
The platform is hosted at app.agnost.ai; its published plans list self-hosting under the custom Enterprise tier, so ask them what that involves. Its GitHub org (github.com/AgnostAI) publishes SDK code, including a Go SDK under MIT, but the platform itself is not open source as far as its public materials show. smolanalytics is MIT end to end: one Go binary, one data file, self-host free forever with every feature.
What does smolanalytics cost, and can I self-host it?
14-day full trial, no card. Pro is $49/mo for 1M events, Scale is $149/mo for 10M, and extra usage is $8 per million. Self-hosting is free forever under MIT: one Go binary, one data file, no external services. There is no free cloud tier.
Get the verdict, not another dashboard.
Drop one snippet, ask in your editor, own your data. 14-day full trial, no credit card.