Playbooks · Coding tasks
Letterpress-1: a growth agent that files PRs
Written November 7, 2026
Letterpress-1 reads your site and market, then opens pull requests with live previews. No public benchmark, $100 a month in, 5x overage. The honest read.
What it does
Letterpress-1, in beta from The Letter Company, is a coding agent, harness, and IDE for growth engineering. It reads a website and its market, decides what should exist, builds the change in a sandbox, and opens a pull request with a live preview. Nothing lands until you choose a preview, a pull request, or a merge after checks pass. It works in three modes: Build, which creates a page or feature from a prompt; Plan, which turns a goal into a build plan of growth actions; and Fix, which finds site issues and dispatches pull requests to fix them. Every change follows the same loop — read the site, plan, build in a sandbox, review its own change, open the pull request — and each change returns a preview link of your own site before anything merges.
It runs on Letter's hosted platform against your git repository, with no local GPU needed, and Letter says setup takes under five minutes. It comes in six surfaces: the web platform, a Slack app, the letterpress CLI, a website portal, a standalone app, and your own agent — Claude Code, Codex, or any agent with MCP gets the same tools the app is built on. Letter says 760 tools are exposed through one registry over MCP, REST, and CLI. The models are frontier models plus fine-tuned small models, and Letter does not name them. The product page also lists growth primitives — Proofs, Simulations, Benchmarks, Phantoms, Kernels, Letterbases, Visitors, Mailers — positioned as replacements for tools like G2, Semrush, Profound, RB2B, and Instantly.
What it costs
Letter bills one shared token pool, and a token is a cent at list price. Agent runs are priced at what the models cost plus 20%, typically 150 to 900 tokens per pull request, or $1.50 to $9 at list. A published article costs 1,030 tokens. Every account starts with 5,000 free tokens, 6,000 on a work email, with no card.
| Plan | Per month | Tokens | Overage per 1,000 tokens |
|---|---|---|---|
| Basic | $100 | 10,000 | $50.00 |
| Starter | $500 | 50,000 | $50.00 |
| Growth | $1,000 | 130,000 | $38.46 |
| Scale | $2,500 | 400,000 | $31.25 |
| Enterprise | $10,000 | — | $25.00 |
The overage rate is the pricing trap, and it belongs up front, not in a footnote. On Basic and Starter, overage is $50 per 1,000 tokens — five times the $10 list rate. Work keeps going at the overage rate up to the spend cap you set; at the cap, Letter stops starting new work and asks your admins for more. Letter's own pricing page says moving up a plan is usually cheaper than living in overage. Unused tokens do not roll over: a plan's tokens refill every month and the unused ones do not carry over, with one exception for a free account's unused tokens when it adds a card.
The arithmetic is easy to get wrong, so run it before paying. Letter's pricing page estimates 20 pull requests a month at 500 tokens each as 10,000 tokens — the entire Basic pool, before a single article, phantom site, or visibility answer. A team shipping daily will live in overage on Basic within the first week.
What is unproven
There is no public benchmark for Letterpress-1. Letter shows one side-by-side page build against Codex — the same prompt on the same site, both runs unedited, with the Letterpress-1 page in the site's own design and Codex's in a clean template look. That comparison is Letter's own, shown on Letter's own page. It is not an independent test, and this page does not present it as one. The quality claims — more tasteful change, brand-ready copy, visuals maintained automatically — are Letter's claims about its own product, with no independent measurement behind them in the material behind this page.
What to do before paying
- Start on the free tokens. 5,000 free tokens, 6,000 on a work email, is enough for a handful of pull requests. Use them on a real page you care about, and judge the pull request it opens — the diff, the preview, the copy — not the demo on Letter's homepage.
- Set a spend cap first. Every plan includes a spend cap and alerts. Set the cap before the first paid run, not after the first overage bill.
- Count pull requests, not months. Estimate your monthly pull requests and multiply by the 150-to-900-token range. If the total lands near the plan's pool, the next plan up is cheaper than the overage — Letter says so on its own pricing page.
- Keep a human on the merge. Nothing lands until you say so, and the preview exists so you can read the change before it merges. Review the pull request the way you would review a colleague's: the agent's taste is unproven, and the merge is yours.
What can go wrong
- Paying $100 and spending $250. A Basic pool covers about 20 average pull requests. Everything past the pool bills at five times the list rate, and work keeps going until the spend cap stops it. The monthly bill is the pool plus the overage, not the pool alone.
- Losing unused tokens every month. Tokens do not roll over. A quiet month does not bank credit for a busy one, so a plan sized for a busy month wastes tokens in a quiet one.
- Trusting the Codex comparison. One side-by-side build, chosen and shown by Letter, proves Letter can stage a good demo. It proves nothing about the next fifty pull requests on your site.
- Assuming the models are named. Letter does not name the frontier models or the fine-tuned small models. Any claim about which model runs your pull request goes beyond what Letter publishes.
Limits
- This page reads Letter's product and pricing pages and the MarkTechPost launch write-up. It reports no Letterpress run here — no account, no paid plan — so it says nothing firsthand about pull-request quality, speed, or cost on a real site.
- All prices come from Letter's pricing page. The page carries them as published; a price Letter changes after publication is not reflected here.
- No public benchmark exists for Letterpress-1. This page states that plainly rather than filling the gap with the vendor's comparison.
Related playbooks
- GitAgent: gated commit messages and PRs, locally
- Audit apps that installed agent plugins for you
- BotPsy 0.8B: an agent model that fits on-device
- Claude agents now dream: memory that consolidates
Sources
- Introducing Letterpress-1 — Letter
- Pricing — Letter
- The Letter Company Releases Letterpress-1: A Coding Agent Built for Growth Engineering — MarkTechPost
Back to: Coding playbooks hub · Also: Cheapest way into an always-on agent · Plan allowances: what each agent plan lets you do · Model prices: what the major models cost