AI agent for Basecamp.

He lives in your projects, not in another tab.

Invite Tentmate to Basecamp

No Basecamp account yet? Grab a free one first.

Mention him in a comment. Assign him a to-do.

He answers like anyone else on the team.

He works in the projects you invite him to.

Bound by the mechanisms you already know.

He works while you sleep.

Long-running sessions, deferred and perpetual tasks.

Invite Tentmate to Basecamp

No Basecamp account yet? Grab a free one first.

This is what it looks like

A question from somebody in the project, and an answer that ends in something they can open.

Pitch: the billing rework

Jonas on Sep 27 · Notified 6 people

Tentmate turn this pitch into a project. Six-week appetite, starts Monday.

Everything you need is in the pitch above.

Tentmate, AI agent

Reading the pitch and its two attachments…
Breaking the appetite into shippable scopes…
Creating the project, assigning the kickoff…

Project is up, structured the way the pitch is scoped.

What I took from it: three scopes, so three to-do lists rather than one flat pile: metered usage, the plan switcher, the invoice email. Each one can be finished on its own, which is what makes the six weeks answerable.

What I left out: the proration rabbit hole is written down as a to-do marked no-go rather than folded into a scope. It is recorded, and it is out of everybody's way.

Everyone on the pitch's notified list is on Billing rework, Cycle 3 already, and the kickoff to-do is assigned to you, due Monday.

Q3 discovery: eleven calls

Priya on Sep 26 · Notified 4 people

Tentmate read all eleven transcripts and turn them into something we can act on.

Not themes. I can find themes myself.

Tentmate, AI agent

Reading all eleven transcripts…
Counting what was raised unprompted…
Drafting the list, one citation per item…

One finding worth acting on, and it is not the obvious one.

Import is where people stop: eight of eleven, and not one of them called it hard. They called it unclear whether it had worked.

So the first fix is a confirmation screen, not a rebuild. What I cannot tell you is whether it costs you money: nobody in the eleven had churned, because the recruiting was from active accounts.

Every item in Q3 discovery: what to do cites the calls it came from, so you can argue with the evidence rather than with me.

Landing page for the Hiker plan

Marta on Sep 25 · Notified 3 people

Tentmate build a first pass in Paper. The brand doc is attached, type scale and palette are in there, follow them.

Three sections is plenty for a first look.

Tentmate, AI agent

Reading the brand doc: type scale, palette…
Building hero, cards and pricing in Paper…
Pulling the live prices from Billing…

First pass is in Paper. Hero, the three-card block, pricing.

What I followed exactly: the type scale as written, and the four brand colours. Nothing invented.

Where I went off the doc: there is no dark surface token in it and the page is dark. I derived one from the darkest neutral and left it as a comment on Hiker: landing, first pass, rather than quietly adding a colour to your palette.

The prices are pulled from the Billing project rather than typed in, so they are the current ones and they stay that way.

Invoice PDF is blank on annual plans

Dan on Sep 26 · Notified 5 people

Tentmate this reproduces every time on annual and never on monthly. Screenshot attached.

Two customers have now asked twice.

Tentmate, AI agent

Reproducing it on an annual subscription…
Tracing the renderer to the null interval…
Opening a pull request with a failing test…

Found it, and there is a pull request open against it.

Cause: the renderer reads the billing period off subscription.interval, which is null on annual subscriptions created before the March migration. A null interval renders the empty template instead of throwing, which is why nothing was ever logged.

Fix: fall back to the plan's own interval, plus a backfill for the 214 subscriptions in that state. #4471, blank invoice PDF on annual plans carries a test that fails on main and passes on the branch.

Not merged and not deployed. The backfill touches live rows, and somebody should look at it before it runs.

Last week

Sam on Sep 27 · Notified 7 people

Tentmate what actually happened last week? Stripe and PostHog, together.

Not a dashboard screenshot.

Tentmate, AI agent

Pulling last week and the four before it from Stripe…
Matching the trials against PostHog…
Looking for a deploy inside the window…

Revenue is up 6% on the week, and it is one customer rather than a trend.

The 6% is a single annual upgrade on Tuesday. Take it out and the week is flat against the four before it. Nothing else moved in Stripe.

The thing worth your attention is not revenue: trial-to-paid fell from 22% to 14%, and in PostHog the whole drop sits in trials that began on the pricing page. Trials from the docs converted at their usual rate.

What I could not determine: whether the pricing page changed. There is no deploy in that window I can see, so I can tell you the behaviour moved and not why. The queries are in Week of Aug 4: Stripe and PostHog if you want to check my working.

Export is missing rows

Ellis at Northwind on Sep 27 · Notified 4 people

Tentmate our export came up short again. The CSV has 1,247 rows and the report above it says 1,248.

Second time this month. Is it something we are doing?

Tentmate, AI agent

Reproducing the short export…
Finding where the tail is being lost…
Opening a pull request, queueing the re-runs…

Not something you are doing. I reproduced it here, and it is worse than one row.

What happens: the export writes in batches of 500, and the last part-batch is flushed after the file has already closed. Any export whose row count is not a multiple of 500 loses its tail. Yours lost one row. It can lose up to 499.

The fix: flush before close, with a test that exports 501 rows and counts them. It is up for review as #4488, flush the final export batch, and it is not deployed yet, so I am not going to tell you it is solved.

Both of your short exports are queued to re-run the moment it ships. Nothing on your side to change in the meantime.

[data-thread-step] { opacity: 1 !important; } [data-thread-panel] { visibility: visible !important; } /* Stacked panels with no way to switch between them would overlap into mush. */ [data-thread-panel] + [data-thread-panel] { display: none; }

Why shared agent

Why managed agents beat everybody running their own harness locally.

Shared memory

Nobody starts from scratch

A local agent forgets when the session ends, and what it does remember is stuck on one laptop. What he works out in a Tuesday to-do he still knows in someone else's thread on Friday.

An agent per job

Focused context

Every discussion, to-do item, and chat get its own isolated context, which prevents context rot and allows agents to focus on the task.

Granular access

Only what each project needs

Each agent hears only the Basecamp projects it is invited to, and reaches only the MCP servers you gave it. The agent in the client project can never reach Stripe.

Parallel work

As many at once as you need

Mention him in six threads and six runs start at once. Nobody waits behind anybody else's job, and nothing is waiting for one laptop to be awake.

Anyone can ask him

No terminal, no install, no keys

Once he's set up, asking him is typing a name in a comment. Designers, ops and clients get the same teammate the engineers configured, without configuring anything.

Audit and improve

You can see how he's used

Every run is logged: who asked, what he ran, how long, what it cost. Read the ones that went wrong, fix the skill, and the whole team gets the better version.

One price for the whole company

Everyone in the company shares one subscription, and it is the only AI the company needs. Start free, move up as you ask more of him.

Start here

Camper

$19 /month

A daily assistant for a small team.

1,500 credits a month

Agents: 1
GitHub repositories: 1
MCP connections: 3
Subagents: Not included

Get 500 free credits

No credit card required.

Hiker

$49 /month

A workhorse for a busy team.

4,500 credits a month

Agents: 2
GitHub repositories: 3
MCP connections: Unlimited
Subagents: Unlimited

Try with 500 free credits

No credit card required.

Climber

$149 /month

Enough of them to run the company.

15,000 credits a month

Agents: Unlimited
GitHub repositories: Unlimited
MCP connections: Unlimited
Subagents: Unlimited

Try with 500 free credits

No credit card required.