Home / /uses
/uses · the actual stack behind Mama

$ cat ~/.what-we-actually-use
The boring honest answer to "what do you use?"

This page is the developer-tradition /uses: the hardware, the editors, the services, the libraries, the fonts. The actual ones — not the aspirational ones. Updated when something changes, not when something gets sponsored.

First published: Jan 2026 · Last revised: May 19, 2026 · 11 sections · 56 items · ~ 6 min skim
01Hardware 02Editor & CLI 03Code stack 04Infrastructure 05Data & models 06Design & type 07Team ops 08Go-to-market 09Desk 10Habits 11What we skip
01 / hardware

Hardware.

5 picks · updated when someone replaces theirs

Same laptops across the team — keeps the development environment from being a snowflake per engineer. The headphones rule is the only one we'd defend in court.

MacBook Pro 14" · M3 Pro · 36GB
apple.com/macbook-pro
Standard issue for every engineer. Same chip, same RAM, same screen, so when a build breaks on one machine, it breaks on all of them. We reimburse for the larger battery upgrade — it pays back in airport hours.
since 2025 $2.4k / seat
Would we switch? No, not until the next Apple Silicon generation lands. Linux-on-laptop is good now but the maintenance tax is real for a small team.
LG 5K2K UltraWide 40"
lg.com · model 40WP95C
Reimbursed for everyone, optional. About 7 of 11 use one. The ones who don't tend to be people who prefer a 16" on its own — we don't argue.
since 2024 $1.6k
AirPods Pro (Gen 2)
apple.com/airpods-pro
The headphones rule. Reimbursed. Transparency mode + the case auto-pause is the actual feature — it's what made async pair-programming feel survivable.
since 2024 $249
Would we switch? Tried Sony WF-1000XM5s. ANC is technically better. Latency on calls was worse. Stayed.
Sony A7C-II (for video)
sony.com
One camera, shared. For changelog videos, customer interviews, and the rare big-deck talk. We don't have an in-house video team — this is owned by whoever's doing the recording that week.
since 2026 $2.0k
Shure SM7dB (for podcast / videos)
shure.com
The microphone matters more than the camera. Bad audio is the fastest way to make polished-looking video feel cheap.
since 2026 $499
02 / editor & cli

Editor & command line.

8 picks · most-changed category on the page

Each engineer picks their own, but we standardize on the shared config files so opening a teammate's project doesn't lose two hours to setup. Cursor is our most-debated pick.

Cursor
cursor.com
Our default editor. About 8 of 10 engineers. The "tab-completion against the actual codebase" is the real win — not the chat panel. We pay for the Business tier so prompts don't train on other teams' code.
since late 2024 $40/seat/mo
Would we switch? Open question every quarter. Zed and the Continue.dev + VS Code stack are both viable. We'd need a quarter of bake time before committing.
Ghostty
ghostty.org
Terminal of choice. Replaced iTerm for most of us in early 2026. The GPU-accelerated rendering is the thing — we run with a lot of tail logs open.
since Jan 2026 free
fish shell
fishshell.com
Default shell for everyone new. Autosuggestions out of the box — saves the dotfile-debugging tax that comes with zsh + oh-my-zsh. People keep their existing zsh if they prefer.
free
mise
mise.jdx.dev
Runtime version manager. Replaced nvm + pyenv + asdf with one tool that reads a single mise.toml per project. New-laptop setup went from a half-day to about 20 minutes.
free
Raycast
raycast.com
App launcher + clipboard history. The window-management extensions replaced Rectangle for most of us. Pro tier is paid for the team.
$10/seat/mo
1Password (CLI & app)
1password.com
Secret store + SSH key vault. The op CLI integration with our .env loading was the moment we stopped having "wait, what's the staging DB password" Slack messages.
$8/seat/mo
Cleanshot X
cleanshot.com
The screenshot tool. Reimbursed for every team member. The pinned-screenshot feature and quick annotation get used dozens of times a day. Worth more than its price.
$29 once
Linear
linear.app
Bug tracker + project tool. No Jira, no Asana, no Notion-based "task DB." Linear handles all engineering work and most of the cross-functional work too. Cycles are 2 weeks.
$14/seat/mo
03 / code stack

The code stack.

8 picks · the boring foundation

Most of these are conservative on purpose. We're building a moat in the data layer; we're not trying to be exciting in the framework layer. Boring stacks ship faster.

TypeScript
typescriptlang.org
Backend and frontend. Strict mode, no implicit any, no // @ts-ignore in main. The build fails on warnings.
free
Next.js + React (App Router)
nextjs.org
The web app. Server components for the data-heavy pages, client components where they earn it. We use a thin RSC pattern — not a "let's port everything to server" pattern.
since rewrite Q4 2025
Would we switch? Probably not. Watching Remix / TanStack Start closely. Migration cost is hard to justify against shipping product.
Hono + Bun (services layer)
hono.dev
Internal services. Things like the brief generator workers and integration ingest are small Hono apps deployed independently. Cold-start time is the killer feature.
free
Drizzle ORM
orm.drizzle.team
Type-safe queries. Replaced Prisma in early 2026. The generated SQL is closer to what we'd hand-write and the migration story is cleaner for our schema.
free
PostgreSQL 16
postgresql.org
Primary OLTP database. Hosted on Neon for the app, self-managed on a single beefy RDS instance for the model orchestration metadata.
since day 1
ClickHouse Cloud
clickhouse.com
Analytics warehouse. All signal events, all brief generation telemetry, all the read-heavy product analytics. The decision to buy this is on /how-we-decide.
usage-based
Tailwind CSS
tailwindcss.com
The styling layer. Combined with a thin set of component primitives we wrote ourselves — no Radix, no shadcn import. The component layer is small enough that owning it is cheaper than adopting one.
free
Biome
biomejs.dev
Lint + format. Replaced ESLint + Prettier with a single binary that runs in < 1 second on the whole repo. The pre-commit hook stopped being a thing engineers complained about.
free
04 / infrastructure

Infrastructure.

7 picks · everything we wouldn't build ourselves

All infrastructure follows the rule: if it isn't on the moat, buy it. We deploy on Vercel for the front, Fly for the long-running workers, and lean on managed services everywhere else.

Vercel
vercel.com
Hosts the Next.js app. Preview deployments per PR is the single most valuable feature for our review process.
~ $400/mo
Fly.io
fly.io
Long-running workers. Brief generators, the model orchestrator, ingest workers. Multi-region for the data-residency edges.
~ $1.2k/mo
Neon (Postgres)
neon.tech
Primary app database. Branchable Postgres — every PR gets a database branch. Made the staging-data-divergence problem disappear.
~ $600/mo
Cloudflare (DNS, R2, queues)
cloudflare.com
DNS, asset storage, message queue. R2 for the saved-brief PDFs, queues for the signal-ingest pipeline, DNS for everything.
~ $250/mo
Resend
resend.com
Transactional email. Magic links, daily digests, and Slack-mirror outputs. React Email templates means the marketing site and product emails share components.
$90/mo
Sentry
sentry.io
Error tracking. Server, client, and worker errors all flow here. Connected to the on-call rotation.
$280/mo
BetterStack (Logtail + Uptime)
betterstack.com
Logs + uptime monitoring. The uptime page that backs /status is theirs. Cheaper than DataDog for our scale and the UI doesn't fight us.
$200/mo
05 / data & models

Data & the model layer.

5 picks · this is the moat

The model orchestration is the thing we'd build in-house ten times before buying. Vendor model APIs we rent. Everything that decides which model to call and how to combine outputs is ours.

Self-hosted model orchestrator
internal · "loom"
The actual moat. Picks which model for which step of a brief, handles caching, fallbacks, and quality scoring. ~14k lines of TypeScript, written by us, with the test suite as version-controlled fixtures.
in-house
Anthropic Claude (Sonnet, Opus)
anthropic.com
Primary reasoning model for brief generation. Sonnet for the bulk of the work, Opus for the harder sections that need more depth. Bills via the orchestrator.
usage-based
OpenAI (embeddings, gpt-4 fallback)
openai.com
Embeddings for signal clustering. Also a fallback reasoning model — the orchestrator routes here on rare cases where the primary times out or returns low-confidence.
usage-based
Voyage AI (reranker)
voyageai.com
Reranks retrieved signal candidates before composition. Materially better than naive cosine ranking for our use case. Tested against three alternatives before picking.
usage-based
Turbopuffer (vector store)
turbopuffer.com
Vector storage for the signal corpus. Picked over Pinecone and Qdrant for the cost-per-million curve and the cold-start behavior — we have a lot of seldom-queried tenants.
usage-based
06 / design & type

Design & type.

5 picks · the brand layer

The visual identity is owned by the founding designer who started in May. Most of the design tokens live in code; Figma is where we explore, not where we ship.

Inter (UI & marketing)
rsms.me/inter
The primary typeface. Free, neutral, designed for screens. The custom feature flags (cv11, ss01, cv02) are part of the design system — they soften the geometry slightly.
free
JetBrains Mono (monospace)
jetbrains.com/mono
Mono treatment everywhere. Tags, prices, metadata, terminal blocks. The ligatures are turned off — they're distracting for the kind of label-y monospace use we do.
free
Figma
figma.com
Design exploration + light prototyping. We don't hand off final designs from Figma — the design system lives in code. Figma is the sketchpad.
$15/seat/mo
Lucide icons
lucide.dev
Icon set. The strict 24×24 grid keeps icons visually consistent. We extend it with about 20 custom signal-glyph icons in the same style.
free
Loom
loom.com
Async screen recording. Used for design reviews and customer-facing walkthroughs. The transcript feature pays for itself.
$15/seat/mo
07 / team ops

Team ops.

6 picks · how 11 people stay coordinated

Small team, so we lean on a small number of tools used heavily, rather than a wide tool surface. The rule: a tool earns its place when it removes a recurring conversation.

Slack
slack.com
Real-time, project channels, customer Slack Connect. About 18 internal channels — we keep the surface small on purpose.
$15/seat/mo
Notion
notion.so
Long-form docs and the internal handbook. Not a task tracker, not a wiki-database. Pages live or die by whether someone updates them.
$10/seat/mo
Google Workspace
workspace.google.com
Email, calendar, docs, sheets. The boring obvious choice. SSO source-of-truth for everything that supports Google login.
$18/seat/mo
Linear (again, but for company-wide work)
linear.app
Marketing, ops, customer-facing work all live here too. The single-tool-for-all-tasks bet has held up.
$14/seat/mo
Cal.com
cal.com
Booking pages. Open-source, customizable, plays nicely with Google Calendar. The white-labeled embed is on /demo.
$15/seat/mo
Mercury (banking)
mercury.com
Company banking. The bill-pay flow and corporate-card controls are why we stayed.
free tier
08 / go-to-market

Go-to-market.

5 picks · we use our own product first

We use Mama on Mama. The accounts we research are companies that look like our ICP — outbound teams, sales consultancies, RevOps leaders. Dogfooding the product is the first GTM line.

Signal Mama
signalmama.com (yes, ours)
Our outbound runs on Mama. About 200 researched accounts a month, mostly the outbound-team and consultancy ICPs. The product roadmap is heavily shaped by what frustrates us when we use it ourselves.
in-house
HubSpot (CRM)
hubspot.com
Our own CRM. Eats our own integration. Helps us catch real bugs from the inside.
~ $500/mo
Customer.io
customer.io
Product emails (digests, in-app nudges). Picked over Loops and Resend Audiences for the campaign-branching tools.
$150/mo
Plausible (analytics)
plausible.io
Marketing site analytics. No cookie banner needed. The dashboard fits on one screen. Anything more is product analytics, which lives in our internal ClickHouse.
$19/mo
Stripe
stripe.com
Payments + billing. Stripe Billing for the per-account metered overages. Stripe Tax for the indirect tax pieces.
% of revenue
09 / desk

The desk.

4 picks · physical environment, reimbursed

Remote-first means everyone's desk is the company's desk. We reimburse up to $1,500 per new hire toward chair, desk, light, and one indulgent thing of their choice.

Herman Miller Aeron (or Steelcase Leap)
hermanmiller.com
Chair tax. Either of these is approved without conversation. Anything else needs a sentence of why.
~ $1.0k
Uplift V2 standing desk
upliftdesk.com
The standing desk default. Bamboo top, programmable height presets. Nothing fancy, just reliable.
~ $650
BenQ ScreenBar Halo
benq.com
Monitor light. The single best ergonomic purchase for screen work. We push every new hire to add this before anything else.
$199
Yeti coffee mug
yeti.com
One indulgent thing. Reimbursed at hire time. Some pick a plant. Some pick a beautiful pen. Some pick a Yeti.
$40
10 / habits

Habits.

6 picks · not tools, but how the tools get used

The tool list above is only half the story. These are the operating habits — the things a new hire would notice in their first two weeks and find worth keeping.

01 No-meeting Wednesdays
No meetings allowed on Wednesdays. Async only. This is the day shipped features get shipped.
02 Weekly memo (every Friday)
One memo from the founders, posted in Slack and archived to Notion. Covers: what shipped, what didn't, what we learned, what we're worried about. It's the source-of-truth for "what's going on this week."
03 Pre-meeting prep doc
Every meeting longer than 30 minutes has a written agenda posted at least an hour before. No agenda = meeting cancels.
04 Customer call on every team's calendar
Engineering, design, marketing — everyone joins at least one customer call per month. Reading the recording doesn't count.
05 "Walk it" days
Once a quarter, everyone takes a full day to use the product from scratch and write down what's broken. Output is a single Linear cycle of fixes. The bar is "would a new customer notice this," not "is it a P1."
06 PRs reviewed in < 4 hours
Median PR-to-review time is under 4 hours. The norm holds because everyone reviews; it's not a "senior engineer" job.
11 / skip

What we skip.

6 picks · the absence is the signal

/uses pages always brag about what's in. The more useful list is what we deliberately don't use — and the reasoning. Most of these are not value judgments about the tool; they're judgments about our shape.

· Jira
Used it once, escaped. Linear's fewer-knobs design fits our team size. Would revisit at 50+ people, not before.
· Salesforce as our CRM
We integrate with it; we don't use it ourselves. At 11 people the implementation cost would be higher than the value.
· A separate "growth" stack
No Segment, no CDP, no martech triangle. Plausible + Customer.io + the ClickHouse warehouse covers everything we measure. Adding more would create coordination cost without revenue.
· Kubernetes
Our deployment surface is small enough that Fly + Vercel cover everything. K8s is the right answer eventually. It is not the right answer at 11 people.
· A microservices architecture
Modular monolith with a few worker services. Splitting services is a tax we'll pay only when team coordination cost exceeds deploy coordination cost. Not yet.
· Daily standup
Async written check-in in Slack instead, posted before noon. Standups become performance theater above ~6 people. We never made it work.
If you've made it this far

The point of /uses isn't the tool list. It's the transparency.

If you're sizing us up — as a vendor or as a place to work — you now know the actual ground truth of how we build. That should make every other claim on this site more believable, not less.

If you spot something we should swap, or a tool you'd want us to consider — write back. We update this page when the stack moves, not when a vendor pays us to.

Try Mama → Work here