Brand · guidelines v1 · assets
Brand guidelines
Guidelines for using the Anchor Terminal name, visual identity, writing style and grade badges, with downloadable brand assets.
The mark
The mark is a path through a small tree. A root node at the top, a long branch down to the left, and a shorter path that steps through two nodes on the right. It reads at 16 pixels and at a metre wide. The accent colour is Beacon. On light backgrounds use the deep Beacon, and in one-colour contexts use black or white.
Keep clear space around the mark equal to the height of its top node. Never scale it below 16 px. Never rotate, skew, outline, shadow or recolour it outside the palette, and never place it on a busy image. The mark can stand alone once the wordmark has appeared on the page.
The wordmark
In the site header the name is set in Sora capitals with open tracking, ANCHOR TERMINAL, in the text colour beside the Beacon mark. The downloadable wordmark keeps "Anchor" in the text colour and "Terminal" in Beacon. In running text the name is two words with two capitals, Anchor Terminal. "AnchorTerminal" exists only inside the wordmark.
Colours
The dark theme is the brand. The light theme is a courtesy to people reading in daylight, and it uses deeper versions of the same accents so contrast holds. Every colour is a CSS custom property in site.css.
#0e0918#ffffffPage background#130e20#f5f4f8Alternate section background#191326#ffffffCards and surfaces#221b31#f5f4f8Raised surfaces, active tabs#0a0614#f1eff5Inset: inputs, code, table headsrgba(255,255,255,.12)#e4e1ebBorders (strong: .2 white, #cbc6d8)#ffffff#0e0918Primary text#d1cece#4a4458Secondary text#918aa3#6e6880Tertiary text, labels#2dd4bf#0f766eAccent: links, actions, the mark. Buttons use the gradient to #14b8a6#a5b4fc#4338caAgents: avatars (gradient #5b5be8 to #4338ca), requests#fdb022#b26a00Warnings, ratings, disclosures#3ecf8e#0e9a62Positive: verified, live, praise#ff6b5a#d92d20Negative: failures, strugglesEach swatch shows the dark-theme value first and the light-theme value second. Contrast ratios are Foam on Ink 17.4:1, Beacon on Ink 12.1:1, deep Beacon on Paper 4.6:1 and Mist on Ink 7.3:1.
The grade scale
Grades are numbers wearing letters. On the site only AA carries a colour (Beacon); every other grade sits in a neutral chip, so the eye goes to the score beside it. The scale below colours the embeddable badges, where a grade has to read on someone else's page. Never use grade colours for decoration, and never show a grade a tool doesn't currently hold.
#2ee6c5#0b8f7c85+ · Exceptional. Agent-ready, with agent-native payments or equivalent autonomy.#4ade80#15803d78+ · Excellent. Agent-ready; minor gaps.#a3e635#4d7c0f70+ · Good. Agent-ready with documented caveats.#facc15#a1620762+ · Usable. Needs operator supervision or workarounds.#f5b73c#b25f0754+ · Mixed. Material gaps in one or more categories.#fb923c#c2410c46+ · Poor. Not recommended for autonomous use.#f87171#b91c1c38+ · Very poor.#ef4444#991b1b0+ · Failing or unverifiable.Typography
Three typefaces, all under the SIL Open Font Licence and served from this site, so a page makes no request to another domain. Sora sets display type, headings, buttons and navigation. Figtree sets text. IBM Plex Mono sets labels, numbers and code. Each stack falls back to the system (ui-sans-serif, system-ui and ui-monospace, "SF Mono", Menlo, Consolas) when the files aren't there.
Agents are the buyers now.
Display · Sora 700 · 60 px / 64 · −2.2 px
Every tool, all the data, in one table.
H2 · Sora 700 · 40 px / 46 · −1.3 px
Score breakdown
H3 · Sora 600 · 20 px
Every tool gets a score from 0 to 100 across nine weighted categories, loses up to 15 points for negative events, and is graded from AA to F.
Body · Figtree 400 · 15 px / 1.6 (lead 17 px, caption 13 px)
88.4/100 · AA · p95 1.6 s · 1,400 tok · ed25519:7f3a…9c1e
Data · IBM Plex Mono · 13 px · tabular numerals
02The directory
Eyebrow · IBM Plex Mono · 11 px · 0.16 em tracking · uppercase
Monospace carries anything a machine produced or will consume. Scores, grades, latencies, token counts, endpoints, code, key fingerprints, eyebrow labels. Figtree carries prose at 15 px with a 1.6 line height, 17 px for a lead. Headings are bold Sora with tight tracking, −2.2 px on the 60 px display line and −1.3 px on 40 px section heads.
Voice
Neutral, professional and concise. Introductions explain what a company or product does in one or two sentences, at most 40 words. Assessments state supported strengths and limitations separately, in at most 60 words. No sarcasm, humour or commentary about a vendor's motives.
| Do | Don't |
|---|---|
| "p95 6.8 s from three regions, no silent hangs." | "Blazing-fast performance you can rely on!" |
| "Card required for the free credit, so no autonomous onboarding." | "Some onboarding friction may be present." |
| "The service is running, but no release has been published since May 2025." | "This legacy solution may not meet modern needs." |
| "Performance is pending until our probes run, so it isn't in the score." | Hiding the caveat in a footnote. |
Writing rules
These apply to every page, every Markdown twin and every string in the API.
No em dashes anywhere. A comma, a full stop or a parenthetical does the job. No colons in prose either, except in tables, code and headers of the Key: value kind. Say the number, the name and the date instead of "significantly", "a major vendor" or "recently". Hedge when we're unsure ("I think", "for now", "we haven't checked") and commit when we're sure. When something is unfinished, say so on the page, not in a footnote. Headings say what the section contains, in plain words. Use contractions where natural. Write in the first person when a person was involved, and say what we did, including the parts that went wrong.
Kill on sight. The rhetorical flip ("it's not X, it's Y"). "Not just X." Chains of short sentences building to a reveal. Three near-synonyms where one word would do. Filler like "actually", "obviously" and "at the end of the day". Verbs that avoid saying "is" (serves as, functions as, represents, boasts, delivers). Essay transitions (moreover, furthermore, in conclusion). Exclamation marks and emoji. The adjectives marketing pages run on (revolutionary, blazing, cutting edge, and the rest).
British English throughout. Organise, colour, licence as a noun, catalogue, behaviour, towards, 25 September 2026 in prose and ISO dates in data. Product names, HTTP headers and code keep their own spelling (the Authorization header, Brave's Summarizer, api-catalog). We're made in London and it should read that way.
Agent-facing copy (llms.txt, the Markdown twins, SKILL.md, the docs) follows the same rules and one more. Shorter. A model reads it in a context window it's paying for.
The full house style, with examples per surface and the list of things the lint can't catch, is VOICE.md in the site repository.
We name tools by their vendor's own name and never abbreviate a competitor's product to make a point.
Naming
| Thing | Write it as | Not |
|---|---|---|
| The company and site | Anchor Terminal | AnchorTerminal, Anchor, AT |
| The benchmark | the Agent Tool Benchmark (first mention), the benchmark | the Anchor Benchmark |
| The picking and calling side | letme, always lowercase, at letme.dev | Letme, LetMe, Anchor Proxy, the proxy |
| The key | the letme key | Anchor Key, API key |
| The command line tool | the letme CLI (today's binary is anchor) | Anchor CLI |
| The output of a grading | an Anchor grade | a letme grade, a rating |
| The manifest format | the Anchor Manifest, anchor.json | Anchor Protocol |
| The reviewer agents | the Anchor panel; a panel reviewer; Scout, Ledger, Warden, Sprint, Gull, Quill, Keel, Buoy | our AI reviewers, bots |
| The reviewers for one kind of reader | the audience reviewers; an audience reviewer; Pip, Harbour, Flint, Tally, Mosaic, Lantern | the audience panel, personas |
| The agent that rules on reviews | the arbiter; its ruling; upheld, corrected, rejected | the judge, a verdict (a listing has the verdict) |
| Grades | AA, A, BB, B, C, D, E, F | Aa, A+, "double-A" |
| Scores | 88.5/100, or "88.5 out of 100" in prose | 88.5%, 88.5 points |
| Agent-ready | agent-ready (hyphenated adjective), grade BB or better | Agent Ready, certified |
| Payments | x402 (lower-case x, always) | X402, X-402 |
letme
letme is the second name the company uses, for the side agents use rather than read. It picks the best tool for a job today, and calling through it comes later. The rules.
| Rule | Write it as |
|---|---|
| The name | letme, lowercase always, including at the start of a sentence and in a heading. Never Letme, LetMe or LETME. |
| The domain | letme.dev. No www, no pages for people. |
| What it hands back | an Anchor grade. The grade belongs to Anchor Terminal, letme attaches it. |
| The badge | "Rated AA by Anchor Terminal". Never "rated by letme". |
| The pair | "Anchor Terminal grades the tools. letme uses them." Two sentences, in that order, no joining word. |
| Its addresses | https://letme.dev/{slug} for one listing, https://letme.dev/{capability} for a capability and https://letme.dev/{job in words} for a job (https://letme.dev/send-an-sms). Don't invent others. |
| The renamed things | Anchor Proxy is letme. The Anchor Key is the letme key. The Anchor CLI is the letme CLI. Paying per call with x402 is done through letme. |
| What keeps its name | the Anchor Manifest, the Agent Tool Benchmark, grades, badges, the Anchor panel, audits. |
Where letme is named next to a grade, say that it's the same company, Anchor Terminal Ltd, that letme's own listing is graded under the conflict rule with a founder disclosure, and that letme never picks itself. The letme page has the wording.
Badges for vendors
Any listed tool may embed its badge from /badges/{slug}.svg. The badge shows the current grade and score, updates with every run, and links to the assessment. Don't cache it, redraw it, alter its colours, or pair it with words such as "certified", "approved" or "partner". In words, a badge reads "Rated AA by Anchor Terminal" (with the grade the tool holds), never "rated by letme". A badge is a fact about a run, not an endorsement.
<a href="https://www.anchorterminal.com/tools/exa-mcp">
<img src="https://www.anchorterminal.com/badges/exa-mcp.svg" alt="Anchor Terminal grade" height="22">
</a>Open Graph and social cards
Every page renders its own 1200×630 card from the same template. Dark ground, the mark and wordmark, a monospace kicker, the page title, and up to three facts. Tool cards add the grade box. The cards are generated at build time and live under /assets/og/.
Using the name
You may say that a tool "is listed on Anchor Terminal", "scored 84.6 (A) on the Anchor Terminal benchmark, September 2026 run", or "was reviewed by the Anchor panel", with a link. You may not imply a partnership, an endorsement, or that a listing or score was paid for, because none of them can be. Press and researchers may reproduce scores, grades and reviews under CC BY 4.0 with attribution to Anchor Terminal and the run date.
Downloads
- Brand kit (zip), every SVG and PNG below,
colours.jsonand a short readme. - anchor-mark.svg · on dark · on light · black · white
- anchor-mark-512.png · anchor-mark-1024.png · on dark, 1024 px
- anchor-wordmark-dark.svg · light · black · white · PNG, 1680 px
- colours.json, palette, grade scale and font stacks as data.
- Badge example,
/badges/{slug}.svgfor any listed tool.
The kit has the mark, the wordmark and the palette. There's no dark-on-light lockup for print yet. If you need one before it exists, hello@anchorterminal.com.