PostHog Slack app
Contents
The PostHog Slack app is in beta. Commands, scopes, and behavior are still moving – expect some rough edges and reach out in Discord if something looks off.
@PostHog brings PostHog into any Slack channel, thread, or DM. Mention the bot with a question about your product data and it answers in-thread using the PostHog MCP. You can also send the bot a direct message – no mention needed – for a private one-on-one conversation.
Mention @PostHog with a fix, an edit, or a feature idea and it plans the work in a sandboxed environment, edits files, runs your checks, and opens a draft PR. The PR links back to the Slack thread it came from, so reviewers can find the original context. It's the same agent that powers PostHog AI and PostHog Desktop, summoned to where your team already chats.
See task environment detection if your scripts or agent instructions need to detect a PostHog task in a cloud sandbox.
Before you can mention the bot you'll need to connect the integrations:
- A Slack workspace connection for each project.
- GitHub, to ship code. A personal GitHub connection lets the agent open PRs as you. Without one, a task can fall back to your project's GitHub App installation and open the PR as the PostHog bot.
Ask about your product data
Ask the bot a question about your product data and it answers in the thread. It can query anything in your PostHog project, including insights, SQL, session replays, errors, and feature flags. You don't need a connected repo for this.
For trends, funnels, and other results that are easier to read as a chart, the agent adds a chart under its reply. Reply in the thread to ask a follow-up or change the breakdown.
Dig into a subscription or report
To ask about an insight or dashboard subscription delivered to Slack, reply in its thread and mention @PostHog. The agent reads the subscription message, so you don't need to say which insight or dashboard you mean:
Anyone in the channel can ask follow-ups in the same thread. This also works for other messages PostHog posts to Slack, such as Self-driving inbox notifications.
Start a task
Mention the bot with what you want done:
You can ask for things that mix code and product data because PostHog Desktop has your PostHog project wired up, so feature flags, experiments, events, and live metrics are all in the agent's context:
If you have a single repo connected, the task runs against it immediately. If you have several, name the repo (org/repo or a GitHub link) in your message or the thread, or let the agent pick one. Your routing rules guide that pick. If the pick fails, the bot opens a repo picker in-thread, with a No repo needed option. The picker times out after 15 minutes. After that, re-mention the bot to try again. See how the agent picks a repo for the full order.
While the agent works, the thread shows a progress message with a link to the agent logs. When it opens a pull request, it posts a Pull request opened card, and later a card when the PR is merged or closed. If the PR was opened as the PostHog bot, the card says so. See Connect your personal GitHub. Once the agent finishes, it posts a final summary of its work to the thread. If a run fails, the bot posts a Task Failed message. Reply retry in the thread to continue from the last checkpoint.
Replies from the agent carry a footer. It links to the session in PostHog, names the model used and its reasoning effort, and offers a Configure link into the App Home tab. The footer also names the project the run belongs to, when the person who started the thread can access more than one project connected to the workspace. There are also thumb icons to rate the answer.
When the agent's answer asks you to pick an option or approve a next step, it can add up to three reply buttons under the answer. Click one to send that reply in the thread as yourself.
Follow-ups in a thread
Reply in the thread to send a follow-up to the running agent. Anything you say while the run is active is forwarded as a new message in the same conversation, so you can steer mid-run ("actually, skip the test files") or answer a question the agent asked.
Once the agent finishes, you can reply to it by sending a new message in the thread with @PostHog to continue the existing conversation.
Teammates can send follow-ups too. Each message is forwarded to the running agent as a new turn in the same conversation, so teammates can steer the run together, drop in extra context, or answer a question the agent asked.
Who can join is up to whoever started the thread. Set it in the App Home tab under Multiplayer:
| Option | What happens |
|---|---|
| Let anyone join (default) | Anyone who mentions @PostHog in the thread is picked up |
| Ask me once per thread | The first time someone else joins, you get a prompt only you can see. Approve once and anyone can join that thread |
| Ask me for each person | You approve each new person separately |
| Ignore them | Only you can talk to the agent in your threads |
Follow-ups are multiplayer: each turn runs as the PostHog user who sent it. For GitHub, it depends on how the run started:
- Runs that started with a personal GitHub connection. Before each turn, the agent switches to the sender's own GitHub connection. Commits on that turn carry their name, not the name of the person who started the thread. If the sender has no personal GitHub connection, the agent can't fetch, push, or open PRs on their turn, and the rest of the conversation carries on as normal.
- Runs that fell back to the project's GitHub App installation. Every turn uses the installation, whoever sends it. Commits and PRs come from the PostHog bot. If the person who started the thread connects their GitHub later, their next turn switches the run to their identity.
Reply without tagging the bot
Once a task is running in a thread, the bot can also pick up replies that don't mention @PostHog. Whether it does is up to whoever started the thread – set your preference in the App Home tab under Thread follow-ups:
| Option | What happens |
|---|---|
| Ask them first (default) | If the bot thinks a reply is meant for it, it asks first with a prompt only the replier can see |
| Pick them up automatically | If the bot thinks a reply is meant for it, it answers without asking |
| Leave them alone | Untagged replies are ignored – mention @PostHog to reach the agent |
The setting applies to every reply in threads you started, including your own. The bot is deliberately conservative about what counts as "meant for it" – teammates talking to each other are left alone.
Questions in channels
The bot can also offer to answer a question you post in a channel without tagging it, when your PostHog data can answer it. By default it asks first, with a prompt only you can see: click Yes, answer it and it replies in a thread. Change this in the App Home tab under Questions in channels (Answer automatically, Ask me first, or Never answer). The setting only covers your own top-level posts in channels.
Rate an answer
Agent replies carry a thumbs up and a thumbs down. Click one to rate it. A thumbs down opens a short What went wrong? box where you can tell us what you expected instead.
Reacting with a 👍 or 👎 emoji on the agent response also counts as rating, but it won't open the What went wrong? box.
A rating is feedback for us, not an instruction for the agent. It changes nothing about the run in flight. The reply is left as it is, so the rest of the thread can't see how you voted. Anyone who can read the reply can rate it. The click itself is what records the rating, so closing the box without typing still counts.
Ratings from Slack sit beside the ones from PostHog AI in the app and from PostHog Desktop. A thumbs down reads the same wherever you leave it.
DM the bot
You don't need a channel to talk to the bot – open a DM with PostHog and message it directly. Everything that works in a channel mention works in a DM: data questions, coding tasks, follow-ups, and attachments. It's a good place for quick tasks (or embarrassing questions) you'd rather not spam a channel with.
A new DM opens as a Slack AI assistant panel, with a short greeting and a few suggested prompts to start from. You don't need to mention @PostHog in a DM – just type.
The bot introduces itself by DM: once to whoever installs it, and once to each new teammate who joins your Slack workspace.
Group DMs work too – start one with the bot and your teammates as members, and mention @PostHog to reach it. What you can't do is mention the bot in an existing DM it isn't part of: it can't see or join those conversations.
Commands read the same in a DM as in a channel: /posthog project <id>. See the command reference.
Pick a model
Tasks run on Claude Opus 5.5 by default. To use a different model, just ask for it as part of your message:
Any model available in PostHog Desktop works, and you can request a reasoning effort the same way ("do this with max effort"). The bot reads the request out of your sentence, so shorthand is fine. "Sol", "codex sol", and "opus" all resolve to a model.
A model named as the subject of the work doesn't switch anything. "Add Fable to the model picker" is a task about Fable, not a request to run on it.
To change the model for everything you run, set your default in the App Home tab under AI model. You can also set it in PostHog Web under Settings → Environment → Task agents. They are the same setting: the Home tab edits your PostHog task agent default, which applies to Slack, the task composer, and PostHog Desktop alike. Project admins set a project-wide default in the same place.
The precedence is: model named in the message → auto model choice (if you turned it on) → your PostHog default → the project default → Claude Opus 5.5.
You can also switch models mid-run from a follow-up message, with one constraint: the runtime (Claude or Codex) is fixed once an agent starts, so moving to a model from the other family requires starting a new thread. For example, you can go from Opus to Fable mid-run, but not from a Claude model to a Codex one. Each reply's footer shows the model and reasoning effort actually used.
Auto model choice
Turn on Use auto model choice in the App Home tab under AI model and PostHog picks the model for each new task. It's off by default.
A fast model reads your first message and picks the cheapest model likely to get it right on the first try:
- Stronger model: new features, refactors, bugs to track down, unclear asks that need digging, plans, and reviews.
- Faster model: quick questions, single queries or summaries, and small edits like a typo or a version bump.
It judges the work, not the message length. It stays on your usual runtime, and when two models fit equally well it picks your default. A model you name in your message always wins. If the pick fails, the task runs on your default, and follow-ups keep their task's model.
Pick a project
When your Slack workspace is connected to more than one PostHog project, the bot routes each mention to one of them. Name a project in the message that opens a thread and it answers from there:
Only the first message of a thread can pick a project. A task, its thread, and its run all belong to one project, so a follow-up can't move it – start a new thread instead.
Everything else is a default you set once: your personal project in the App Home tab under Project routing or with /posthog project <id>, and the workspace-wide fallback by a Slack admin. If you have no default yet, the bot asks you to pick a project and remembers your choice. Pick a project covers the full precedence.
Attachments, files, and canvases
Attach files to your message and the agent reads them: images, PDFs, logs, CSV exports, and other files. Up to 15 attachments per message, 30 MB each (10 MB for PDFs). Files posted earlier in the thread come along too, up to 15 more, with 50 MB in total per turn. Files over a limit are skipped, and the bot tells you which ones.
Reading attachments needs the files:read Slack scope. If the bot says it's missing, ask a project admin to reconnect the Slack app.
Paste a screenshot of a broken page, a design, or a chart, then tell the agent what to do with it:
Send a stack trace, a log file, or a CSV export the same way, in place of a wall of pasted text.
What the agent sends back
Most answers are text, but the agent can also send:
- Charts – a chart image under the reply, with a button to open it in PostHog.
- Canvases – a Slack canvas for longer documents, like reports or plans. Ask for changes in the thread and the agent updates the same canvas.
- Files – a file uploaded to the thread, such as an image (PNG or SVG), a spreadsheet (XLSX or CSV), a PDF, a Word document, an HTML page, or a ZIP archive.
Canvases and files need the canvases:write and files:write Slack scopes. Slack doesn't add new scopes to an existing install, so if you get text instead of a canvas or file, ask a project admin to reconnect the Slack app.
The App Home tab
Open the PostHog app's Home tab in Slack (or click Configure in the footer of any bot reply) to manage your setup in one place:
- Workspace activity – tasks, PRs opened and merged, merge rate, median run time, models used, and most active teammates, over the last 7, 30, or 90 days. Visible to Slack workspace admins and owners only.
- Project routing – your personal default project, plus the workspace-wide default (editable by Slack admins).
- AI model – which model and reasoning effort your mentions run on, and where that comes from, plus the Use auto model choice option. Editing it here edits your PostHog task agent default, so the change also applies in the task composer and PostHog Desktop.
- Thread follow-ups – how the bot treats untagged replies in threads you started.
- Questions in channels – whether the bot offers to answer your untagged channel questions.
- Multiplayer – who can join threads you started.
- Channel welcome – where the greeting goes when someone adds the bot to a channel. Visible to Slack workspace admins and owners only.
- Linked accounts – your PostHog account link and personal GitHub connections.
- Tasks – the tasks you've started from Slack with thread activity in the last 14 days, filterable by repo and status, with links to the web, desktop, and PR.
Emoji cues
The bot reacts to your @PostHog message as it works through your request, so you can scan a busy channel and see what's still in flight without opening every thread:
| Emoji | What it means |
|---|---|
⏳ :hourglass: | Your message is queued behind another one in the same conversation |
🔍 :mag: | Picking which repo to run the task against |
👀 :eyes: | The agent is working on your message |
❌ :x: | Something went wrong |
Link unfurling
When you paste a PostHog link in a channel the bot is invited to, Slack shows a rich preview with the title, type, and description. This works for insights (with a chart image), dashboards, notebooks, feature flags, experiments, surveys, cohorts, error tracking issues, Self-driving reports, support tickets, and tasks.
The bot checks that the person who pasted the link can access the item. If they can, the preview is posted to the channel, so everyone in the channel sees it. Links are never unfurled in Slack Connect channels. See Slack link unfurling for details.
Related
- Setup – connect Slack and personal GitHub
- Commands – full command reference and limitations
- Pricing – how Slack app usage bills
- PostHog Slack app overview – the product page
- PostHog Desktop – the desktop app for longer-running, multi-repo agent work
- Slack integration in PostHog – subscriptions, link unfurl, generic Slack setup
- Slack as a Data Pipeline destination – send event-driven messages