Agentry is a product data layer for your AI coding agent. It gives Claude Code, Cursor, Codex, and other agents one queryable place for errors, analytics, deploys, answer contracts, and project context so the agent can investigate, report, automate, and fix from your editor.

How is Agentry different from Sentry?

Sentry is built around a human-facing dashboard that engineers open to triage errors. Agentry is built around an AI coding agent that triages and fixes errors directly in your editor. Both capture errors; Agentry adds analytics and deploys, has no SDK, and lets the agent do the investigation work that humans normally do in a dashboard.

How is Agentry different from PostHog or Mixpanel?

PostHog and Mixpanel are product-analytics tools built for human analysts to explore data via UI. Agentry is the agent-facing data layer: your agent asks the questions, writes the HogQL, reads errors and deploys alongside analytics, and can publish or generate the output you need.

Does Agentry require installing an SDK in my app?

No. Agentry has no SDK by design. Your AI agent generates a small fetch helper at install time that POSTs runtime errors and analytics events from your app, while CI or provider post-deploy automation POSTs deploy events after successful releases. You can use plain curl, fetch, or any language with an HTTP client. Total install footprint is well under 50 lines of your own code.

Which AI coding agents does Agentry work with?

Any AI coding agent — Claude Code, Cursor, Codex, Cline, Windsurf, and others. Install the Claude plugin with `claude plugin install agentry@agentry`, drop the skill into other agents, or just point any agent at https://api.agentry.sh. Nothing runs server-side; your agent calls the HTTP API directly.

What does Agentry cost?

Free during the current beta. Planned launch tiers are Free at 50k events/month with 90-day retention, Pro at $39/month for 1M events/month with 180-day retention, and Scale at $149/month for 10M events/month with 365-day retention. Every plan includes every feature; only event volume and retention vary. Billing is not wired yet and requires explicit opt-in when it ships.

How do I install Agentry?

Tell your AI agent: 'Install https://agentry.sh/install.md for me'. The agent fetches the install handbook, starts a device-flow login (you sign in at agentry.sh/cli with GitHub, Google, or magic-link email), mints an API key, creates a project, plans the business funnel and signal map, instruments errors and analytics in the app, wires deploy capture from CI/provider post-deploy automation, and only reports success after verify-install passes.

your app

events

Agentry

prompts

output

Product data layer for your AI agent.

Error monitoring, product analytics and deploy attribution in one place your agent can investigate, query and build on.

Install agentry.sh/install.md for meInstall https://agentry.sh/install.md for me

Agent will onboard itself and then your app

Works with any AI coding agent

How it works

Everything is just a prompt away.

Send the data, then ask. Your agent does the work — engineering, growth, customer success.

Your app

Sends runtime signals on meaningful actions — logs, analytics, revenue. CI/provider hooks send deploys after release.

Agentry

data layer

Stores everything. Queryable via SQL. Webhooks for automations.

Output

Your agent builds you dashboards, fixes bugs, automates reports, what ever you ask for.

Growth / PM

Find the magic activation moment

~14s · 4 tools · writes a local /admin/* page

claude — ~/myapp

> What action in the first 24h after signup correlates highest with users still being active at day 30?▎

On it — let me run the 60-day cohort retention diff and write you a local /admin/activation-funnel page you can open in your browser.

⏺ GET /v1/projects/:id/event-names (min_count: 100)

└ 12 events: signup_completed, integration_connected, …

⏺ POST /v1/projects/:id/analytics/query (cohort_retention, 60d)

└ top 5 first-24h actions ranked by day-30 retention

⏺ POST /v1/projects/:id/cohorts (activated_users)

└ cohort_id: co_a8f2 · matching users: 142

⏺ Write (src/pages/admin/activation-funnel.astro)

└ 84 lines · uses your existing Tailwind theme

✓ Wrote src/pages/admin/activation-funnel.astro

→ localhost:4321/admin/activation-funnel

localhost:4321/admin/activation-funnel

Activation moment found

integration_connected → 78% retention

First-24h actions · day-30 retention · 60d cohorts

vs baseline

4.3×

integration_connected 142 users · 78%

team_member_invited 98 users · 71%

first_query_run 320 users · 64%

$pageview baseline 620 users · 18%

tutorial_completed below baseline 84 users · 16%

⚠ negative signal

Users who finish the tutorial retain below baseline. The tutorial appears to be substituting for engagement, not feeding it.

API calls used

GET /v1/projects/:id/event-names
POST /v1/projects/:id/analytics/query
POST /v1/projects/:id/cohorts
Write (local .astro page)

You get

·A reusable activated_users cohort
·A local page at /admin/activation-funnel
·Reusable in feature flags & surveys
·Works from any AI coding agent

See more Growth / PM examples →

Engineering

Triage silent bugs while you sleep

set up once · runs nightly · drafts PRs when safe

codex — ~/myapp

> Wire up a nightly Routine at 2am: pick the top silent bug, investigate, draft a PR if the fix is <20 LOC. Slack me each morning.▎

On it — I'll dry-run against your case backlog, register the Slack webhook, schedule the routine via Claude Code Routines, then drop a local /routines/rt_nightly/preview page so you can see what tomorrow's post will look like.

⏺ GET /v1/projects/:id/cases (dry-run, silent bugs only)

└ 7 candidate bugs (≥5 users affected, no triage yet)

⏺ POST /v1/projects/:id/webhooks (slack → #eng-mornings)

└ webhook_id: wh_eng_morn · signing_secret: agws_…

⏺ /schedule (Claude Code Routine)

└ rt_nightly armed · cron: 0 2 * * * UTC

⏺ Write (src/pages/routines/rt_nightly/preview.astro)

└ 62 lines · preview of tomorrow's #eng-mornings post

✓ Wrote src/pages/routines/rt_nightly/preview.astro

→ localhost:4321/routines/rt_nightly/preview

localhost:4321/routines/rt_nightly/preview

agentry-bot → #eng-mornings

tomorrow at 08:15 · preview

queued

Drafted overnight

PR #420 — null-guard in BillingForm.tsx:118

Affected 7 users · Size 14 LOC · tests pass typecheck clean

draft fix(billing): null-guard in BillingForm #420

-  const street = user.address.street;
+  const street = user.address?.street ?? '';
+  test: renders fallback when address is null

API calls used

GET /v1/projects/:id/cases
POST /v1/projects/:id/webhooks
/schedule (Claude Code Routine)
Write (local preview page)
gh pr create (overnight, in the routine)

You get

·A routine that fires at 02:00 UTC nightly
·Slack post in #eng-mornings before your coffee
·Draft PR when the fix is small and safe
·Triage note when it isn't — no surprise commits
·You sleep through the bug-triage rotation

See more Engineering examples →

CS / RevOps

Predict churn 30 days before it happens

~22s · 5 tools · writes a local /admin/* page + draft email

cursor — ~/myapp

> Find behaviors that predict churn. Compare users who churned last month vs users who renewed. Build a saved cohort of at-risk users I can email.▎

Got it — I'll build two cohorts (April churners + renewers), diff their pre-decision behavior, save a live at_risk_users cohort, then write a local /admin/churn-risk page with the chart + a draft save email.

⏺ POST /v1/projects/:id/cohorts (churned_april)

└ 87 users matched

⏺ POST /v1/projects/:id/cohorts (renewed_april)

└ 412 users matched

⏺ POST /v1/projects/:id/analytics/query (behavior_diff, 60-30d window)

└ 4 high-signal events · comment_added −95%

⏺ POST /v1/projects/:id/cohorts (at_risk_users, live)

└ 31 paying users match · auto-refreshes daily

⏺ Write (src/pages/admin/churn-risk.astro)

└ 104 lines · chart + draft save email

✓ Wrote src/pages/admin/churn-risk.astro

→ localhost:4321/admin/churn-risk

localhost:4321/admin/churn-risk

Strongest signal

comment_added drops 95% before churn

Built from April · 87 churners vs 412 renewers · 60–30d before decision

At-risk now

31

comment_added strongest signal renewers 8.2/mo · −95%

team_member_invited renewers 0.31/mo · −94%

first_query_run renewers 14.8/mo · −86%

$pageview renewers 38.7/mo · −84%

★ insight

Collaboration goes to zero first — content events follow. Churn is predictable 30 days out from behavior, not from billing signals.

at_risk_users · live cohort auto-refreshes daily

To: paddy@startup.io

Subject: paddy, are we still useful?

Hey paddy — noticed your team's usage dropped over the past month. That's usually one of two things: you've outgrown us, or something's in the way. Either way I'd love to hear from you — what changed?

API calls used

POST /v1/projects/:id/cohorts ×2 (churners + renewers)
POST /v1/projects/:id/analytics/query
POST /v1/projects/:id/cohorts (live at_risk_users)
GET /v1/projects/:id/users/:distinct_id/summary
Write (local .astro page)

You get

·A live at_risk_users cohort, refreshes daily
·A drafted save email per at-risk account
·A local page at /admin/churn-risk
·Lead indicator (comment_added) — 30 days early
·Hook up Resend / SendGrid to auto-send

See more CS / RevOps examples →

None of these are Agentry features. Your agent builds whatever you ask on top of the data. See more examples →

Why this exists

The interface changed.

Modern observability tools were designed for humans clicking dashboards. AI agents change the interface.

Agents can:

▸ build customized views
▸ automate operational workflows
▸ generate integrations
▸ analyze trends and suggest improvements
▸ investigate incidents and fix bugs

Agentry is designed from the ground up for agentic software development.

Features

What's in the box.

Built for AI agents as the primary user, not humans clicking dashboards.

ONE EVENT STORE

Logs, errors, analytics, and deploys land in the same project. One key, one dataset, one query surface.

NO SERVER-SIDE MAGIC

Agentry stores; your agent transforms. Every workflow is driven by the HTTP API and reviewable in your repo.

CASES & SUPPRESSION

Duplicate errors collapse into one case by fingerprint. Your agent records suppression rules — noise teaches itself out.

DEPLOY-AWARE DEBUGGING

Automatically correlate regressions with releases and commits. One tool call answers "what shipped before this broke?".

AGENT-GENERATED OPS

Queries become conversations. Dashboards become customized artifacts. Ask, your agent runs the SQL and writes the page.

WEBHOOKS & AUTOMATIONS

Signed webhooks on any event. Cron jobs your agent writes. Automate fixes, reports, and alerts — your code runs, not ours.

FAQ

Questions you might have.

Migrating from Sentry?

Yes. Send error-shaped JSON to /v1/logs/ with Authorization: Bearer ${AGENTRY_DSN}. Agentry accepts the familiar event fields (event_id, exception.values, stacktrace.frames), but setup stays explicit and native.

Migrating from PostHog?

Yes. Post JSON events to /v1/analytics/ with the same event, distinct_id, and properties shape. PostHog remains the analytics engine under the hood; Agentry gives your agent install context, query blueprints, answer contracts, and deploy context over HTTP.

Do I need to use Claude Code?

No. Agentry is just an HTTP API, so it works with any AI coding agent — Cursor, Windsurf, Cline, Codex, or your own. We optimize for Claude Code because that's what we use, but nothing's tied to it.

Do I need to install anything?

Almost nothing. Point any agent at https://agentry.sh/install.md for setup and https://agentry.sh/agentry.md for reference — it drives everything over plain HTTP. Your agent can also grab an AGENTS.md / Cursor / Windsurf adapter from https://api.agentry.sh/adapters. No SDK and no Agentry client package.

What languages does this support?

Anything that can POST JSON. There is no SDK to install — your agent generates a 25-line fetch helper at install time, tuned to your stack. Reviewable in 30 seconds, no vendor dependency to vet, no upgrade cycle.

Does the server transform my data?

It stores, retrieves, and queries — and runs deterministic transforms like sourcemap unmangling, which return the release and source they used so you can reproduce the result. No server-side LLM, no opaque magic.

What about humans who want a dashboard?

Your agent writes the dashboard as a real page in your repo. You can commit it, edit it, and ship it. No SaaS UI to learn — and no permanent dashboard config rotting over time.

Do you train AI on my data?

No. We never run LLMs server-side. The agent that reads your data is your Claude Code or another agent runtime running on your machine. Your data stays in your Agentry project.

Agentry isn't trying to replace Sentry's source-map polish or PostHog's cohort-analysis depth. We sit on top of both worlds with a single ingest, a schema-shaped output, and a workflow that ends in a PR instead of a notification.

Get started