by The Letter Company
Coding agent that turns site and market signals into pull requests
Built for Git-based SaaS teams that want SEO, GEO and landing-page work delivered as reviewable pull requests. It replaces ad hoc prompting of a general coding agent for site changes. The team is six people and the product is a beta with no published security page, so check data handling before connecting a repository.
Delivers growth work such as SEO, GEO, landing pages and A/B tests as pull requests with live previews, including edits made in place on the live site with Ctrl+L.
The pricing page, read on 15 November 2026, bills sites, brands, mailboxes and tracked visitors from one pool, which raises the cost of leaving, but every output is a standard pull request that any coding agent could also produce, which keeps that cost low.
Main trade-off
Beta software that publishes no security or privacy documentation yet.
Six monthly token pools: Hobbyist $25, Basic $100, Starter $500, Growth $1,000, Scale $2,500 and Enterprise $10,000. Custom plans are quoted on request. Agent runs are billed as model cost plus 20 percent, converted into tokens, and a sandbox minute costs 2 tokens.
| System | Type |
|---|---|
| GitHub | Native |
| GitLab | Native |
| Slack | Native |
| MCP server | API |
Letterpress-1 entered beta in November 2026 as an agent that reads a site, drafts a change inside a cloud sandbox and files it as a pull request with a live preview. Its scope is growth engineering: SEO, GEO (visibility in AI answers), landing pages, A/B tests and visitor identification.
Build, Plan and Fix modes
Build makes what you ask for and returns a preview, Plan proposes actions toward a goal such as GEO or SEO, and Fix finds broken pages and opens a pull request for each issue.
Ctrl+L editor on your own site
Press Ctrl+L on your site, double-click any text, change it and the edit becomes a pull request; a link ending in #letter opens the editor on phones too.
Cloud sandbox and CLI
The letterpress command-line tool plans and builds in a cloud sandbox, attaches local files with @path and opens a pull request on GitHub or a merge request on GitLab.
Approval and spend controls
Nothing lands until you pick a preview, a pull request or a merge after checks pass, and a spend cap with alerts stops new work at the limit you set.
Eight growth primitives
Phantoms test GEO in a sandbox, Proofs hold signed customer evidence for agents, Seers watch competitor sitemaps, and Kernels, Traces, Ledes, Mailers and Shredders cover writing, citation tracking, email personalisation and sending.
API and MCP for other agents
Any agent can drive Letter through REST, the CLI or an MCP server at app.letterstory.com, using the same functions the app uses.
| Agent | Where it runs | Price | Maturity | Rating |
|---|---|---|---|---|
| Cloud · API | Per token | Beta | 3.0 | |
| Cloud · Your machine · API | Free plan | Enterprise | -- | |
| Cloud · Your machine | $19/mo | Available | 3.5 | |
| Your machine · API | $5/mo | Beta | 3.4 |
Letter sells one token pool on six plans: Hobbyist at $25 for 2,500 tokens, Basic at $100 for 10,000, Starter at $500 for 50,000, Growth at $1,000 for 130,000, Scale at $2,500 for 400,000 and Enterprise at $10,000. A token lists at $10 per 1,000 on the first three plans, then $7.69, $6.25 and $5.00 on the larger ones. Work past the pool runs at $50, $38.46, $31.25 or $25 per 1,000 tokens until the spend cap you set, and unused tokens do not roll over.
Every account starts with 1,000 free tokens, 1,200 on a work email and up to 2,000 for connecting your tools, and no card is needed. The pricing page lists no recurring free plan, so once the starter tokens are spent you move to a paid plan. A free account's unused tokens move onto its plan when you add a card.
No. According to its npm readme, the letterpress tool does not touch files on your machine. It drafts the change in a cloud sandbox, then opens a pull request, or a merge request on GitLab, against the branch you are on once that branch is pushed. Local files can be attached to a prompt with @path, and uncommitted or unpushed work is flagged because the sandbox cannot see it.
Only if you choose that. Letter's pages say an edit can merge on its own once its checks pass, or wait for your review, and that choice is made in Settings. Until then the work sits as a preview or an open pull request. Daily checks, auto-merge, a spend cap and alerts come with every plan.
The launch coverage on MarkTechPost compares it with Codex, Claude Code and Cursor. Letter's own pitch is that a website reacts to outside signals such as agent traffic, Google rankings and competitor changes, which general coding agents were not built to act on. The vendor reports it orchestrates Claude and Codex models with its own fine-tuned small models; no independent test of its output quality was found as of 15 November 2026.
Letter is a coding agent for growth engineering from The Letter Company, a New York City startup founded in 2026 and part of Y Combinator's Fall 2026 batch, according to YC's directory. Its agent, Letterpress-1, reads a website, plans a change, builds it in a cloud sandbox and opens a pull request with a live preview, so a growth fix reaches the team as ordinary code review. The vendor calls it a coding agent and an IDE for growth work. It counts as an agent here because it plans and carries out multi-step site work by itself and publishes nothing until you approve the result.
You ask in plain words, in the web app, in Slack with @Letter, in the letterpress command-line tool, through the API and an MCP server, or with the Ctrl+L editor on your own site, where you change text on the page and get a pull request. Letterpress-1 has three modes, Build, Plan and Fix, which cover making a page you describe, mapping the site actions for a goal like SEO or GEO (visibility in AI assistants' answers), and sweeping every page for what is broken. The vendor says it orchestrates models such as Claude and Codex together with its own fine-tuned small models. Its command-line tool leaves your local files alone, with builds running in the cloud against the branch you have pushed on GitHub or GitLab.
It fits growth engineers, developer-marketing leads and founders who own a site's Git repository and want SEO, GEO, landing pages, A/B tests and visitor identification handled as reviewable pull requests. Beyond the agent, the platform adds eight building blocks the vendor calls primitives, among them Phantoms for testing GEO in a sandbox, Proofs for signed customer evidence that agents can check, and Seers that watch competitor sitemaps. Hokai files it under coding agents.
Launch coverage dated 2 November 2026 reports Letterpress-1 in beta, and the letterpress command-line tool reached version 0.2.0 on 13 November 2026, after its first npm publish on 20 October 2026.
Skip it if your site does not live in a Git repository you control, or if you need a security attestation: no security, privacy or terms page was found on letter.dev on 15 November 2026. If you only want a general coding agent, Claude Code, OpenAI Codex and Cursor work on any codebase, and launch coverage compares Letterpress-1 with exactly those three. If your goal is only to track how AI assistants mention your brand, Profound measures AI search visibility without writing any code. The decision: pick Letter if you own a Git-hosted marketing site and want SEO and GEO work as pull requests; pick a general coding agent for everything else; and do not connect a repository until you have asked the vendor for its security and data-handling terms if your company requires them.
Autonomy, guardrails, integrations and pricing on this page are read from The Letter Company's own documentation. The verdict is Hokai's reading of them, not the vendor's claim.