What’s new

What we shipped, newest first — one post per feature, and what it means for you. For SDK version numbers and changes that move data you have already looked at, read the SDK changelog.

Subscribe by RSS

Platform

Leaving an organization now ends its CLI and MCP access

When someone leaves or is removed from your organization, the tokens their `ah login` and MCP connections hold for it stop working right away.

Removing someone from your organization in the dashboard took away their dashboard access, but any ah login they had run, and any MCP client they had connected, kept reading your organization's data. Those tokens were checked against our own records, and nothing updated those records when the membership ended. An admin had to find each one on the Tokens page and revoke it by hand.

Now, when a membership ends, we revoke the tokens that person's logins minted for your organization. Deleting a user's account revokes the tokens it minted in every organization. Their access to other organizations is untouched. A connected MCP client keeps working for the organizations they still belong to and notes the one they left.

Tokens created on the Tokens page are not affected. Those belong to the organization, often as a backend's key, and keep working after the person who created them leaves. Revoke one there if it should go too.

Platform

Sessions close on time again

Sessions stopped closing on October 31, leaving session metrics stale; they close normally again, and missed sessions are rebuilt from their events.

If your dashboard looked wrong from Saturday, October 31, this is why. Sessions that had ended stayed open, so anything worked out when a session closes stopped updating: duration, bounce rate, engagement, goal conversions and returning visitors, across every project.

A session closes about 30 minutes after its last event, in a background job. That job asked for all waiting sessions in a single request. When a slow run let enough of them pile up, the request grew too long to be accepted, so nothing closed. The next run had even more to ask for, so it could not recover on its own.

The job now works through waiting sessions in small batches, oldest first, so a backlog can no longer block it. A new automated test covers this case.

No events were lost, and event collection, SDKs and usage metering kept working throughout. Closing a session works from its original events, so the sessions caught in this show the right numbers once the backlog clears. There is nothing you need to do.

Platform

Dashboard forms work again

A dependency update made the dashboard reject its own forms for a few days, so billing, tokens, settings and ah login all hit an error page. They work again.

If you tried to change something in the dashboard this week, you probably saw "Oops! An unexpected error occurred." Pages loaded normally, but pressing the button did not work. That covered upgrading a plan, creating an API token, editing project settings and authorizing the CLI with ah login.

The cause was a security update to the web framework we build on. Its check that a form really came from our own site got stricter and started comparing https as well as the address. Our servers sit behind an encrypted edge and see traffic as plain http, so every genuine form looked like it came from somewhere else and was refused before it ran.

We now tell the check exactly which site is ours. Forms posted from anywhere else are still refused, and a new automated test fails if this ever breaks again.

Nothing you tried went through, so anything you submitted while this was broken was not saved. Please try it again. Event collection, SDKs and existing tokens were never affected.

$ ah login

SDKs

Time on page now counts Safari and Firefox visitors on privacy-hardened sites

Sites that send a strict Referrer-Policy were losing the last batch of every Safari, iOS and Firefox page visit, and the web tracker now delivers it.

If your site sends Referrer-Policy: same-origin or no-referrer and you have domains set on the project, Safari and Firefox visitors looked strangely shallow. Their pageviews arrived, but their leave events almost never did, so they had no time on page and no scroll depth. On one site, iOS Safari logged 11 leaves for about 2,000 pageviews. On the same pages, Chrome logged roughly 3 leaves for every 4 pageviews.

The cause was the tracker's last send as a page closes or a tab is hidden. Under those policies, Safari (including every browser on iOS) and Firefox strip that request's origin. The domain check exists to stop other sites from posting into your project with your public key, so it cannot accept a request with no origin, and it dropped the batch.

The web tracker now sends that last batch in a way that always carries your site's origin. Nothing to install: every site picks up the new build within minutes. Numbers recorded before today are not backfilled.

Platform

See who's on your site right now

The Overview now opens with a realtime panel — visitors on the site this minute, where they're from, and a running stream of what they're doing.

Until now, "is anyone on the site?" meant opening the Events tail, reading raw rows, and counting visitors in your head. Launch days, ad pushes and newsletter sends deserved better.

The Overview now leads with a Realtime panel that refreshes every five seconds:

  • On site now — distinct visitors with any activity in the last five minutes.
  • A trailing window — switch between 5m, 15m, 1h, 6h and 24h to see visitors, pageviews and events over the stretch you care about, with a per-interval sparkline.
  • Countries and pages — where those visitors are and which pages they're on.
  • Activity — a live stream of pageviews, clicks, scrolls and form submits, each with the visitor's country flag, sliding in as they land.

It counts people, not bots: crawlers, suspected bots, test IPs and server-relayed events stay out, the same rule as the rest of the Overview. Need every raw event and its props? "Full tail" jumps to the Events page.

SDKs

Your form's honeypot field now catches bots in your analytics too

Mark a hidden trap field with data-ah-honeypot and any session that touches it is classified as a suspected bot.

Plenty of signup forms carry a hidden "website" field that people never see and bots fill in. The form threw those submissions away, but the analytics still counted the visit. A crawler that loads your page, scrolls to the bottom in one instant and focuses every input looked like an engaged human, right down to the campaign it came from. Ad-review crawlers that visit just after an ad goes live are the usual example.

The web tracker now treats a focus or change on that field as a bot signal. Add data-ah-honeypot to the field or its wrapper, and the session is classified suspected_bot, out of the default human views and campaign numbers. Fields without the marker count only when they are unreachable by keyboard (tabindex="-1"), hidden from screen readers (aria-hidden="true") and named like a honeypot, such as website or bot-field. A real field never meets all three.

Sessions recorded before this release keep their old classification.

Platform

A home page that shows you the questions, not the feature list

The landing page now opens on the questions you'd type into Claude, plays the explainer straight away, and gets you to the Quick start in one click.

The old home page asked you to scroll through six sections before you could tell whether AgentHog answered the questions you actually have. Readers arriving from a newsletter read the page, opened the FAQ and checked pricing, but never found the way in.

The page now opens on "Just ask Claude." A prompt types out real questions, such as "Where are people dropping off in signup on mobile?" or "Did Tuesday's deploy change anything?", and a click skips to the next one. The Quick start button sits right under it.

The four-minute explainer now plays from our own server instead of YouTube. It starts muted with captions on and has the usual controls, so you can turn the sound on, seek or go fullscreen. Under it you'll find one line on what AgentHog is, three cards on how it works, and the same FAQ as before.

The install command at the top of the old page is now a link as well as a copy button: it copies the command and opens the Quick start, where you use it.

CLI

Imported events keep their exact ids

Event ids in ah events, ah user and ah sessions show now come back as strings, so an imported event's id is the real one.

A backdated event you import with ah track --file gets a long id, past what a JSON number can hold exactly. In --json output and the API, those ids came back rounded. The printed id pointed at no event, and two imported events from the same moment could print the same id, so a script that de-duplicated on id could drop real rows.

Event ids now come back as strings of digits, exactly as stored, in ah events, ah user's last events, ah sessions show's timeline, the API routes behind them and the matching MCP tools. Ordering hasn't changed, and counts are still numbers.

Platform

Keep years of history, and a Scale plan for bigger teams

Free now keeps 90 days of history and Pro keeps 5 years. New Scale plan: $250 a month for 10M events and 7 years.

Thirty days was enough to see whether Tuesday's release moved the funnel, but not enough to compare this quarter with last quarter. Pro's 12 months ran out right when a product started to have a history worth keeping.

The windows are longer now. Free keeps 90 days and Pro keeps 5 years, with no price change. From today your history keeps growing toward the new window. Anything that had already aged out under the old window doesn't come back. The date pickers, ah retention notes and ah usage all show the new numbers.

Scale is a new plan for companies running several products with a whole team in the numbers. It's $250 a month, includes 10 million events, keeps 7 years of history, and allows 100 projects and 100 teammates. After 10 million, events are billed at Pro's overage rates. Subscribers can switch between Pro and Scale on /billing at any time. The base price is prorated, and events already sent that month bill at the old plan's rates.

We updated the privacy policy to list the new retention windows.

Platform

Get started with just your email

Every "Get started" button on the site now asks for your email first, sends the docs and your setup link right away, and lets you skip the questions.

Getting started used to mean six questions before we asked where to send anything, and the site had a different button on every page: "Start free", "Create a free account", "Install the agent skill". Leave halfway through and you got nothing.

Now there is one button, Get started, and it asks for your email first. The welcome email with the docs and your setup link goes out as soon as you submit. It does not wait until you finish anything. The short survey after that is optional: every question has a "Skip to the docs" link, and the last one asks whether you want a call with our technical team. Say yes to book one. Say no, or skip, and you go straight to the docs.

On the home page the form sits at the bottom, under the install transcript. The header's Get started button scrolls you to it.

Platform

Watch AgentHog explained in four minutes

The landing page now opens with a narrated walkthrough of the read → act → read loop, in place of the typed terminal demo.

The old hero was a looping fake transcript. It looked right, but it never said why an agent should be reading your analytics, and it went past too fast to follow.

The landing page now leads with the AgentHog explainer: four and a half minutes on what autocapture records, how bot scoring keeps crawlers out of your numbers, and what it looks like when Claude Code or Codex runs ah digest, spots a funnel drop, ships a fix, and logs it with ah changes add so tomorrow's digest shows the before and after.

It is a click-to-play embed. Until you press play the page loads a local poster frame and nothing from YouTube, so the hero stays fast and cookie-free. Pressing play opens the privacy-enhanced player with captions available. The same video lives on the AgentHog YouTube channel if you would rather send someone a link than a landing page.

Platform

See what's in the explainer before you press play

The landing page video now lists its chapters underneath, and the player itself can jump between them.

Four and a half minutes is a real commitment to make to a video you know nothing about. The hero gave you a play button and a runtime, and nothing about whether the part you actually wanted — the install, the bot scoring, the dashboard — was ninety seconds in or four minutes in.

The explainer now carries its chapters in two places. Under the player there is a roadmap: seven lines from "What AgentHog is" at 0:34 through "How to start" at 4:20, so you can see the shape of the thing before you spend the time. Click any of them and the video starts. The video's own description carries the same chapters, which means the player's progress bar is segmented and labelled — once it is playing, you can jump straight to the part you came for, on the landing page or on YouTube.

The embed is still click-to-play: nothing loads from YouTube until you ask it to.

Platform

From your coding agent to your BI agent

Explore how AgentHog connects product data to agent-written reports, with a setup flow that confirms your answers are saved before continuing.

Your coding agent can help investigate user behavior when it has the data behind the question. The new builder landing page explains how AgentHog captures activity across websites, apps, games, and backend services, and connects identified users' sessions and events. Three illustrative workflows show behavior analysis, onboarding drop-offs, and trial-to-paid conversion as reports your team can read and act on.

The setup survey now waits for the server to confirm that your email and answers have been saved before sending you to the next step. If the save fails or times out, your answers stay on the form and you can retry. A failed save no longer counts as a completed survey. Email validation also runs on the server, and campaign tags follow the handoff to the setup guide or account creation.

Platform

Dark mode, following your system by default

The dashboard, reports, docs and admin pages now follow your OS appearance, with a System / Light / Dark switch in the rail if you want to pin one.

An analytics dashboard is the tab you leave open all day, and until now it was the one white tab in a dark desktop. Reading a report at night meant reading it against a bright page, and there was no way to change that short of a browser extension.

Every signed-in page now has a dark theme: the overview and its charts, sessions, revenue, the mobile page, shared reports, the docs and the admin pages. It follows your operating system's appearance out of the box, so nothing needs configuring, and switches live when the OS does. The three-way control at the bottom of the rail (and in the docs and admin mastheads) pins it to light or dark instead; the choice is stored per browser, the way the OS setting it overrides is per machine. Charts, tooltips and report widgets re-colour with the page, and code samples stay dark in both. Printing a report always uses the light palette.

The marketing pages keep their own look and render light in either setting.

Platform

Dashboards and ah sql no longer compete with your own ingest

Dashboard pages, ah verbs, the API and ah sql now answer from ClickHouse; the slowest pages went from over a second to under 200 ms.

Every tile, ah verb and ah sql query ran on the Postgres that takes each ingest batch. A 30-day scan competed with the writes landing that second, and the mobile page, eighteen queries wide, waited on the slowest: over a second on a large project.

Events have been dual-written to ClickHouse since September 4 and reconciled against Postgres around the clock. Reads moved one payload family at a time, each behind a shadow window that served Postgres, ran ClickHouse beside it and compared the rows — zero mismatches, zero errors — before the flip. At the flip the overview's p95 went from 393 to 99 ms and the mobile page on the largest project from 1,138 to 159 ms. Small projects pay a fixed round trip instead: a tile Postgres answered in single digits now takes 20–100 ms, bounded. Results stay analytical, up to a minute behind ingest.

In ah sql a prop_num sum is now a JSON number, a rejected query carries ClickHouse's wording, a result past 10,000 rows asks for a LIMIT, and the Postgres-only idioms the contract never promised stop working. A backdated import is billed against the month it arrives in.

CLI

Read the mobile page from the CLI, the API and MCP

The /mobile numbers come back from `ah mobile`, `GET /api/v1/mobile` and the `mobile` MCP tool — the same payload the page renders.

Every other dashboard page had a verb: ah traffic, ah funnel, ah revenue print the same JSON the page draws, so an agent can read them and a deploy can be checked by diffing before and after. The mobile page did not. Its numbers were assembled in page code, so the only way to read them was to open /mobile and copy the cards, and nothing could confirm that a change to the queries had left them alone.

ah mobile prints the page as terminal output: the six tiles, then actives by platform, install sources and campaigns, the saved onboarding funnel, auth failures, engagement, the weekly and daily retention grids, social rates, paywall conversion, trial and churn cohorts and revenue by month and tier. --since sets the flow window the tiles and tables read; the retention grids and the month cohorts keep the page's fixed spans. --json is the whole payload, the same object GET /api/v1/mobile returns and the mobile MCP tool hands a chat client. The page itself now calls that payload, so the three surfaces cannot drift.

$ ah mobile --since 30dRead the docs →

Reports

Reports now say when the analysis stopped keeping up

A marker on every chart shows how far the prose beside it actually saw, and a section says so when the numbers have moved on without it.

A report holds two things that age at completely different speeds. The charts re-run their queries and stay current. The sentence above them — "signups dipped after the 8/14 release" — was true the morning someone wrote it, and nothing on the page admitted that was three weeks ago. The live chart quietly vouched for the stale paragraph.

Now every time-series chart draws a vertical line at the date the analysis around it was last written. Everything left of the line is data that reading actually saw; everything right of it arrived afterwards, and nobody has looked. Each section carries its own date, and once the data beneath it has run well past that date the section is marked stale, with a count at the top of the report. Because revisions are append-only and every block is identified by its content, the date is the real one: republishing a report around an untouched paragraph does not make that paragraph look freshly written.

You can say how long your own reading should be trusted — <!-- analysis ttl: 7d --> inside a section, or stale_after: in the front matter — because a launch-day readout and a quarterly note age at different rates, and only their author knows which is which. Older reports carry no marks at all; the next update adds them.

$ ah reports read weekly-growthAnalysis freshness →

Platform

Connect an MCP client once and see every organization you belong to

A login token bundles one token per organization into a single credential, routed server-side the way the multi-org CLI does it.

The ah CLI has held logins for several organizations at once since August; the MCP server that shipped yesterday did not. A bearer token belongs to one account, so a client pasted with the CLI's active token saw that account's two projects and read them as the whole list — the other eighteen simply were not there, and nothing said so.

Now the Tokens page has a Connect an MCP client button. It shows the same checkbox list ah login uses, mints one account token per organization you tick (write scope only where you are an admin), and bundles them into a single login token (ah_usr_…) shown once. Paste that into Claude Code, Cursor or the Messages API and projects lists every project across every organization, each row naming its account. A project key routes to its organization by itself; a name that exists in two is refused with the keys. Account-wide tools take an account argument when more than one organization is connected, and whoami, usage, test_ips_list and reports_list answer for all of them, with a warnings entry for any organization whose token has since been revoked.

Nothing changed underneath: every call still reaches /api/v1 as exactly one organization's token, and an ah_tok_ pasted into a header works everywhere it did before. Revoking the login token revokes every token it bundles.

Platform

The API is also an MCP server, for clients without a shell

Every ah verb is now reachable as an MCP tool at /mcp — same tokens, same scoping, same JSON — for the places a terminal can't go.

If you build a Claude-powered assistant on the Messages API, or work in an editor whose only integration is an MCP URL, or run agents in a sandbox where npm i -g is off the table, you could not ask AgentHog anything: the surface was ah, and ah needs a shell.

Now the same /api/v1 endpoints behind ah are served as a remote MCP server at /mcp. It is stateless Streamable HTTP — one POST per JSON-RPC message, one JSON document back, no sessions, no SSE — and it authenticates with the ah_tok_ token you already have, sent as a bearer header. The tools are the CLI's verbs, one for one: digest, traffic, funnel, sessions_show, revenue, experiments_results, reports_read, sql and the rest, each returning exactly the JSON its --json twin prints. A read token sees the read tools and nothing else; a write token adds track, goals, funnels, changes, flags, experiments, revenue and economy mappings, reports, projects and test IPs — the same bodies ah sends, annotated so a client can ask before a delete. Token-scoped and rate-limited throughout.

Nothing about the CLI changes, and inside a terminal agent it remains the better tool. This is the same API for the places a shell cannot go.

Reports

Publish a report your whole team opens

A Markdown file with live charts in it — you write the analysis once, the numbers keep themselves current.

The weekly numbers used to be a ritual: run ah digest, paste the interesting rows into Slack, write two sentences of context around them. By Tuesday the numbers were wrong and the two sentences were the only part still worth reading.

A report inverts that. It is an ordinary Markdown file where a ```widget fence stands in for a chart, a table, or a stat tile. Your prose is frozen — the judgement you made on Monday stays exactly as you wrote it. Each widget carries a constrained SQL query and a TTL, so it recomputes against the account's live data instead of aging into a lie. Publish it and the whole organization reads it at /reports/<slug>.

Revisions are append-only. history, pull --rev and diff reach any earlier version, so last month's read is still there next to the numbers it was based on — which is the part that actually makes a report worth writing.

There is an agent skill for the whole loop, too: /skill/report/SKILL.md.

$ ah reports create weekly-review --file weekly-review.mdShared reports →

Attribution

Singular postbacks land server to server

Point Singular's Internal BI postback at AgentHog and installs arrive attributed, with no client code involved.

Mobile install attribution lives inside your MMP, and your product analytics lives here. Reconciling the two by hand — exporting campaign names, matching them to sessions on a guess — is the kind of chore that quietly never gets done.

POST /postbacks/singular receives Singular's Internal BI postbacks directly. You configure the endpoint once in project settings, Singular delivers each install verdict server to server, and the campaign lands on the person's session utm_* columns. No SDK call, no app release, and nothing that can be dropped by an offline launch.

Precedence is explicit, per column: a real deep-link value beats the MMP verdict, which beats the install-referrer stamp. An organic answer never erases a referrer that already had a real one. Which means every campaign report — traffic, funnels, revenue by source — just starts including paid installs.

Prefer to attach the verdict from the app instead? The SDKs have setAttribution.

Platform

AgentHog moved to agenthog.io

New home for the product, the docs, the tracker and the API — with the old host still serving every existing install.

AgentHog was living at hog.brightmotion.io, which is a fine internal hostname and a poor thing to say out loud. The product now lives at agenthog.io: dashboard, docs, /ah.js, /api/v1, and the sign-in flow.

Nothing you have installed needs to change. The old host keeps serving /api/v1 indefinitely, existing CLI logins keep the host their token was minted against, and AH_API_URL still overrides everything. ah 0.10.0 makes https://agenthog.io the default for a fresh AH_TOKEN or a first-time ah login — upgrade when it suits you.

The one thing worth doing at some point: repoint your <script src=…/ah.js> tag. It works either way, but a snippet that names the product is easier for the next person to recognise.

Platform

Watch your events arrive, live

A running tail of raw events at /events — the page that answers "is my event firing, with the props I expect?"

Every integration has the same first ten minutes: you add the snippet, click the button, and then stare at a dashboard that aggregates on a delay, trying to work out whether the silence means "not wired up" or "not there yet".

/events is the tail. Trigger something in your app and watch the row land — event name, person, and the full prop payload expanded inline. Server-session events are visible by default here, because webhook relays and backend sends are exactly what people are usually debugging.

It reads the same query as ah events, so the page and the CLI can never disagree with each other about what arrived. It refreshes every 15 seconds — fast enough for the trigger-and-watch loop, and it holds no connection open, so nothing to reconnect when your laptop sleeps.

$ ah events --since 30mEvents →
SDKs

Attach your MMP's verdict from inside the app

setAttribution in React Native 0.8.0 and Capacitor 0.4.0 forwards an attribution callback into the session, exactly once.

If you already have Singular's SDK in the app, its attribution callback fires with the answer you want. Getting that answer into your analytics used to mean sending a custom event and then reverse-engineering campaign reports out of it.

AgentHog.setAttribution({ provider, utm_source, … }) forwards the payload into the session's utm_* columns on the next flush's normal cadence. The hard parts are handled: MMP callbacks re-fire on every launch, so each distinct payload is delivered exactly once and the repeats are no-ops; an undelivered payload survives a crash or an offline launch and rides the first flush of the next one. reset() clears it.

A real deep-link value still wins over the verdict, and passing utm_source: "organic" is safe — it won't overwrite a referrer that already knew better.

Available in @brightmotion/agenthog-react-native 0.8.0 and @brightmotion/agenthog-capacitor 0.4.0. Prefer no app release? Use the Singular postback.

Attribution

One name per ad network, and first touch on every person

utm_source is normalized at ingest, so Meta stops showing up four times — and every person now carries a first and latest touch.

The same network arrives spelled differently depending on which pipe it came down. Meta alone turned up as Facebook, fb4a and apps.facebook.com — three rows in a source report, one actual ad network, and a top-channels list you had to mentally re-add.

New sessions now store one canonical name (meta, google, tiktok, …). Unknown values pass straight through unchanged, and the raw value is kept in landing_params, so nothing is lost. Historical rows are not rewritten: a report spanning this deploy shows a seam at that date, with old and new spellings side by side.

On top of that, attribution is now a property of a person, not just a session. ah user derives first_touch and latest_touch from their session ledger, and ah users --source filters people by where they originally came from — which is the shape you want for "how does the cohort from this channel actually behave".

$ ah users --source metaMobile attribution →
CLI

Send events from the terminal

ah track writes a single event or bulk-imports a CSV — for backfills, cron jobs, and anything with no SDK.

Not everything that matters happens in a browser or an app. A refund processed in an admin tool, a nightly job that closes out trials, three years of history sitting in a CSV export from the thing you're migrating off — none of it has an SDK to call.

ah track <event> writes one event. --person and --email tie it to someone, --prop k=v sends bare numbers and booleans as JSON types (while 0022 and 1.9.2 stay strings, which is nearly always what you meant), and --at accepts a date or a wall-clock time and echoes back the exact instant it resolved to and the rule that fired.

ah track --file events.ndjson — or a CSV with --event-col, --ts-col, --person-col, --prop-cols — imports in bulk, packed under ingest's batch caps. --dry-run prints the row count, date range, event names and a sample payload without sending anything, and bad rows are reported by line number rather than half-imported.

One honest caveat, which the verb tells you itself: ingest has no idempotency key. Run an import twice and you have written it twice.

$ ah track signup --person u_123 --prop plan=proThe ah CLI →
CLI

ah sql is now a contract, not a database connection

A published queryable surface, portable helper functions, bound time windows, and per-query ceilings.

ah sql let you ask anything, which sounds generous until you build something on a column that was never meant to be public and it moves under you. Worse, half of what you'd write was Postgres-specific — fine today, and a rewrite the day the storage engine changes.

ah schema now prints the promise: the readable tables, their documented columns, the helper functions, and what is guaranteed to keep working. Only what it lists is promised — internal columns (IP hashes, user agents, bot signals, raw landing params) are no longer readable at all, so SELECT * on those tables fails on permissions rather than silently handing you data you shouldn't build on.

The helpers are the portable ones you actually reach for: prop, prop_num, has_prop, match, median, p90/p95/p99. --since/--until fill :since and :until as bound timestamps resolved server-side, and constrained sessions are pinned to UTC, so date_trunc('day', …) buckets the same way regardless of where the query ran.

Queries run under stated ceilings — 10s statement timeout, 32MB work_mem, 60/min per token — and exceeding one returns an error that names the limit instead of a mystery hang.

$ ah schemaThe ah CLI →
Platform

Install AgentHog into Claude Code as a plugin

One marketplace command installs all three agent skills — integrate, experiment, report.

The skills were already there — Markdown instruction files served raw from /skill/, meant to be dropped into ~/.claude/skills/. That works, and it's one more curl to explain in every onboarding conversation.

There is now a Claude Code plugin. /plugin marketplace add AnniesAI/agenthog-claude, then /plugin install agenthog@agenthog, and your agent has all three: how to integrate AgentHog correctly into whatever repo it's looking at, how to run an A/B test end to end without fooling itself about the result, and how to write and maintain a shared report.

Not using Claude Code? The curl one-liners still work, and every skill is a plain Markdown file any agent can read. That's the whole point of the format.

Platform

See how your account actually uses the API

Per-token, per-endpoint usage for the CLI and HTTP API, plus a monthly event meter against your plan.

Tokens accumulate. A CI job here, an agent there, a script someone wrote in March — and no straight answer to "which of these is still in use, and what is that nightly job doing to my event count?"

Your account's API usage is now recorded and readable: which tokens are calling, which endpoints, how often. Revoking a token stops being an act of faith.

Separately, /usage shows monthly event volume against your plan's allowance, drawn from a maintained rollup rather than counting the raw events table — so the page loads instantly and the number is at most a few minutes stale. It is monitoring only: nothing is enforced, and the meter is there so a surprise is something you notice in week one rather than on an invoice.

$ ah usageThe ah CLI →
Platform

One person, however they typed their email

Emails are normalized at ingest, so [email protected] and [email protected] stop being two customers.

Someone signs up as [email protected] on the marketing site and logs in as [email protected] in the app. Two identities, two retention curves, one real person — and an LTV number quietly split in half.

Emails are now lowercased at ingest before identity resolution, so the case someone happened to use never decides whether their sessions stitch together. It applies to every path that carries an email: the SDKs, server-side sends, and ah track --email.

Small change; it shows up in every person-level number you have.

Economy

Track your game's virtual currency without instrumenting anything new

Declare which events source or sink currency and get flows, a paid/earned split, ranked sources and sinks — retroactively.

Soft-currency inflation is the failure mode nobody sees coming: the balance curve drifts up for six weeks, the store stops mattering, and by the time it shows in revenue you're retro-fitting a sink into a live game. Catching it early normally means designing an economy taxonomy up front and re-instrumenting the client to match.

It doesn't here. If your purchase and reward events already carry an amount, you map them from the CLI: ah economy map --event <name> --source|--sink. Conventional props (amount, currency, balance) are the defaults and get resolved at map time. Mappings are stored in project settings and applied retroactively over all history — no SDK change, no migration, no waiting a month for data.

The report prints one block per currency, never summed across them: sourced with a paid vs earned split, sunk, net flow with supply growth, and ranked TOP SOURCES / TOP SINKS with event and player counts. --by <dim> breaks flows down by platform, country or cohort; --series adds a per-day table.

And an always-on HEALTH line reports unparseable amounts, sign disagreements, and drift between your flows and the client-reported balance — which usually finds the instrumentation bug before it finds the design one. ah economy suggest proposes mappings from events you're already sending, so the first run takes about a minute.

$ ah economy map --event coins_spent --sinkIn-app economy →