The theme is codeTheme from @hanzo/ui/core: Dracula on dark, GitHub
Light on light, the one every Hanzo surface draws. lib/shiki.ts and each
app's source.config.ts import it rather than naming a theme, so vesper
and github-dark are gone. @hanzo/ui moves to 8.27.83, the first release
that carries it.
A block's code area paints the theme's own ground and ink from the pre
shiki wrote, #282a36 in the dark, rather than the site's panel.
The docs home drew its hero and its eight domain snippets as plain text,
and so did the models page's fallback request. The hero highlights its
three doors in a server part (components/hero.tsx); the snippets and the
fallback are Example props in the MDX. No highlighter reaches the
browser.
cloud computes the catalog in Go and serves it at /v1/pricing, so the
docs apps and the zen-models package read the endpoint directly rather
than a host that forwards to it. The Base page links to the pricing page
a reader can actually open.
Co-authored-by: Hanzo Dev <[email protected]>
A version written into a manifest is written into every manifest that wants it,
which is how eslint came to exist in a hundred and four forms across this estate
and how a package came to resolve twice at run time. The catalog moves the number
out of the manifest: a manifest names the package, the workspace names the
version.
Four are held below their latest because the newer release cannot be used, and
the reason is recorded rather than the version alone: @noble/hashes 2 renamed
every subpath the web3 stack imports, @splinetool/runtime 2 depends on a package
that was never published, @dynamic-labs 5.4.1 the same, and react-resizable-panels
3 points its types condition at the bundle.
Co-authored-by: Hanzo Dev <[email protected]>
Every dependency on a @hanzo/@luxfi/@zooai package now names the same version
across the repo: a catalog entry where pnpm supports one, an exact pin where it
does not, and a range only where a peer genuinely means one.
pnpm itself is pinned too — the estate ran eighteen different versions, five of
them too old to read a catalog.
Co-authored-by: Hanzo Dev <[email protected]>
One app carried the Hanzo shell and twenty did not, so a reader crossing from
base.hanzo.ai to zen.hanzo.ai left the ecosystem and landed somewhere that only
resembled it. Five of those twenty had hand-rolled a footer of their own —
sixteen lines of copyright and a repo link, written five separate times.
Each app now loads the brand tokens before its own sheet, reports through
docs-analytics under its own product name, and renders HanzoFooter from
@hanzogui/shell — the same 6-column ecosystem footer, defined once, that
apps/docs already used.
A footer written inside a layout was chrome pretending to be content, and it is
gone. A footer written inside a page is content pretending to be chrome: those
keep their words and become sections, so each page has one contentinfo landmark
instead of two.
Three manifests listed @hanzo/font twice in the same object.
Builds verified across the four shapes this repo has: a static export, a
next.config.mjs app, a single-page app with no route groups, and one that
already reported analytics. Every one renders exactly one footer.
Co-authored-by: Hanzo Dev <[email protected]>
Thirty-one manifests asked for @hanzo/font ^1.9.1 while the lockfile still
resolved 1.8.0, 1.8.2 and 1.8.12, so pnpm install --frozen-lockfile — which is
every workflow under .hanzo — refused to install anything at all.
The floor is 1.9.2 rather than 1.9.1 because 1.9.1's zen.css carries the
@font-face rules and nothing else: the :root block binding --font-zen-sans and
--font-zen-mono is absent, and a sheet reading either property gets an invalid
value and the OS default without saying so. 1.9.2 publishes them again.
Twelve example layouts imported `ZenSans` from `@hanzo/font/sans`, a name
the package stopped exporting at 1.9.0 — the family is called Zen, so the
binding is `Zen`. Every one of those layouts would fail to resolve the
import against a current install, and the `^1.8.x` ranges they pinned kept
resolving to the old export instead of surfacing it.
Across the cloud, insights, bot, engine, ml and router sites the copy read
"SOC 2 compliance", "SOC 2 ready" and "HIPAA / SOC2 ready"; the vault skill
page and its card led with "PCI DSS Level 1" and "PCI DSS-compliant". Hanzo
holds none of these — no report, no auditor, no attestation letter — and
hanzo.ai/trust already says so, so the docs were contradicting our own trust
page. Vault's architecture is the part that was true and it stays: card
tokenization inside an isolated cardholder data environment, which is what
keeps a customer's PCI scope small. The 99.9% uptime SLA on the cloud stat
row goes too; it was a guarantee with no contract under it.
Co-authored-by: Hanzo Dev <[email protected]>
`hanzo-cli`'s [[bin]] is named hanzo-engine, so the docs' bare `hanzo serve`
named a command that binary does not answer to — and `hanzo` is a different
program, the fabric CLI, whose serving verb is now `hanzo up`.
`@hanzo/docs-mdx` writes the collection modules to its outDir, `.docs/`.
Nothing read them there. Apps read a committed `docs/` copy instead, reached
three different ways:
13 apps `@hanzo/mdx:collections/server` + a webpack alias -> ./docs/*.ts
3 apps `@/docs/server`, straight into the committed copy
2 apps the alias, with no committed copy at all
The two with no copy — bootnode-docs and tasks-docs — could not build:
./lib/source.tsx
Module not found: Can't resolve '@hanzo/mdx:collections/server'
and four more built from a copy that no longer matched the content beside it:
./docs/server.ts
Module not found: Can't resolve
'../content/docs/reference/templates/USER.mdx?collection=docs&only=frontmatter'
That is what committing generated code buys: the build reads a snapshot of
the content rather than the content.
One name, one location, every app: `@hanzo/mdx:collections/*`, resolved to
`.docs/` — the generator's own output, rewritten on every build, gitignored.
`tsconfig` gains the matching `paths` entry so the specifier type-checks
instead of relying on `ignoreBuildErrors`; apps/docs stops overriding outDir
in both `next.config.mts` and `gen:collections`; turbo caches `.docs/**`. The
16 committed mirrors are deleted. Examples already did it this way.
`.source` goes with them — this fork's outDir has been `.docs` for a while,
leaving `.source` in the vite plugin's JSDoc, an example's import map, an
oxlint ignore and `.gitignore` pointing at a directory nothing writes.
Also declares in `globalEnv` the env vars the build scripts actually read
(HANZO_DOCS_SYNC, HIPS_REPO_DIR, DOCS_SECTION, the search vars...). Turbo
does not forward undeclared vars, so `HANZO_DOCS_SYNC=0 turbo run build` had
no effect on the sync it names.
turbo run build --filter='./apps/*' --continue
before Tasks: 33 successful, 40 total
Failed: bootnode-docs, docs, engine-docs, ml-docs, router-docs,
spec, tasks-docs
after Tasks: 38 successful, 40 total
Failed: docs, spec (both for unrelated reasons, fixed next)
(cherry picked from commit 3d11394b092530489b60931e46b4938a88a2bdd5)
Co-authored-by: Hanzo Dev <[email protected]>
Zen and enso are Hanzo's own families. Several surfaces described them by
whose API they imitate, or filed them under a competitor's namespace.
- zen-docs embeddings: drop the "Upstream: text-embedding-3-large by OpenAI"
row that presented zen-embedding as a rebadge; correct dimensions to the
1024 the gateway actually returns.
- zen-docs pricing: Zen was priced as "3x the upstream Fireworks inference
cost", with a claim that every model page discloses its upstream model and
provider. Rates are per model; state that instead. Same framing removed
from platform pricing.
- bot-docs gateway config: "anthropic/zen5" -> "hanzo/zen5"; hanzo is the
provider id Bot actually uses (src/commerce/tier-model.ts).
- zen-docs: curl and the Hanzo SDKs lead; upstream-SDK samples move below a
labelled wire-compatibility section. Anthropic Messages page keeps its
substance, loses the drop-in framing.
- Marketing sites: replace provider/model resell counts used as positioning
with what the product is. Our families lead the model grid; third-party
rows stay, since those integrations are real.
- Neutral placeholder for a proxy key in our own samples.
Provider lists Bot genuinely connects to (LM Studio, vLLM, LiteLLM) are left
alone, as is the zen5 expert-pool table, which is real architecture lineage.
Co-authored-by: Hanzo Dev <[email protected]>
apps/docs imported the same component library under three names — 11 files
from `@hanzo/docs/ui/*` (a pure re-export shell), 10 from `@hanzo/docs-ui`
(the Radix fork), plus 5 hand-copied shadcn files under components/ui.
All of it now resolves to `@hanzo/docs-base-ui`.
The 5 local copies are deleted. badge, hover-card, select and the container
card move into base-ui on `@base-ui/react` — preview-card is the hover-card
analog and Select.Value already takes `placeholder`, so the swap is 1:1 and
introduces no Radix. buttonVariants gains `xs`/`md`/`lg` and `link`, purely
additive, so no existing caller changes. Three Radix packages leave
apps/docs; `presence` and `tooltip` stay because non-shadcn code uses them.
global.css now takes base-ui's preset rather than the Radix fork's. The two
are not duplicates: `neutral.css` defines the colours, `preset.css` carries
base styles plus the `@source inline` safelists. Only `default-colors.css`
declares a `--color-fd-*` value, and it is byte-identical across both forks,
so no colour moves.
`types:check` never ran. pnpm installs the mdx bin scope-stripped as
`docs-mdx`, so `@hanzo/docs-mdx &&` died at `sh: no such file` and took
`tsc` with it — silently, in every app, for `lint` too. Fixed in all 32
scripts. apps/docs reports 107 first-party errors; that is the real number,
unchanged by this commit.
Also unbreaks two undefined references in openapi's server entry that a
large upstream merge left behind: PlaygroundAuthProvider was used but never
imported, and pickSchema/encodeInternalRef were orphaned utilities. Both
crashed the docs export. More remain in that package (renderResponse is a
missing function, not an import) — noted, not guessed at.
Co-authored-by: Hanzo Dev <[email protected]>
Three things the workspace disagreed with itself about.
Versions. apps/docs ran ahead on next 16.2.6 / react 19.2.6 / TS 6, the other
21 apps trailed on next 16.1.6 / react 19.2.3 / TS 5.9.3, and the packages sat
somewhere between with five different @types/react pins. Everything is now on
the newest patch of the line it was already on — react 19.2.8, next 16.2.12,
@types/react 19.2.18, @types/react-dom 19.2.4 — and the TS-5 stragglers join
the TS 6.0.3 the root and every package already used. No majors were invented:
6.0.3 IS the top of the 6.x line, and TS 7 is blocked because [email protected]
declares `typescript: ^5.5.0 || ^6.0.0`.
The duplicate key. next.config.mts declared `transpilePackages` twice in one
object literal, so the later list won and @hanzogui/shell and @hanzo/brand had
never been transpiled — the comment explaining why they must be outlived the
setting it described. tsc had been reporting this the whole time as
next.config.mts(193,3) TS1117; the app's `types:check` never reaches tsc (its
script starts with a bare `@hanzo/docs-mdx`, which is not the bin's name and
fails immediately), so nobody saw it. One list now, both groups in it.
Dead weight. Twelve packages apps/docs declared and no file, config, script or
browserslist field referenced: @radix-ui/react-dialog, @floating-ui/react-dom,
@mixedbread/sdk, @orama/orama, openapi-sampler, react-remove-scroll,
algoliasearch, gray-matter, tinyexec, tinyglobby, @types/three (types for a
package that isn't installed) and shadcn (no components.json exists anywhere in
the repo, so the CLI was never initialized). The ones that are real
dependencies of packages/openapi, packages/core and packages/cli keep their own
copies — pnpm's node_modules is isolated, so nothing lost a resolution.
@orama/core stays: apps/docs/app/static.json/route.ts pulls OramaDocument
through @hanzo/docs-core/search/orama-cloud, which types against it.
Verified against a clean worktree at HEAD, since the working tree carries
gigabytes of untracked synced project docs that CI never checks out:
turbo build, apps + packages 43/52 before, 43/52 after, same 9 failures
docs export 1764 html pages, byte-identical page count
deploy.yml gate index/docs/services present, /docs has nav
tsc --noEmit in apps/docs 325 errors before, 324 after (the TS1117)
vitest 169 passed / 13 failed, both sides
The 9 build failures and 13 test failures are unchanged from HEAD and none are
version-related: 7 sibling apps cannot resolve @hanzo/mdx:collections/server
under `next build --webpack`, zen-docs has a committed docs/server.ts importing
a deleted zen5-coder-flash.mdx, and packages/preview cannot resolve its own
@/lib/source alias under waku/vite.
Co-authored-by: Hanzo Dev <[email protected]>
Two independent defects made every documented 'first API call' fail:
1. FICTIONAL MODEL IDS. `zen-omni` and `zen-coder` appear ZERO times in the
live catalog (api.hanzo.ai/v1/models, 440 models). Every SDK sample -
python, typescript, go, cpp, swift, kotlin - told the customer to call a
model that does not exist, so even a correct request 400s. Corrected to the
real ids: zen5 and zen5-coder.
2. DEAD HOST + STALE PINNED MODEL. llm.hanzo.ai is retired and 301s (a POST
returns the literal string 'Permanent Redirect', not a completion), and
`claude-sonnet-4-5-20250929` is long superseded. Both removed: the host is
gone from customer-facing docs entirely rather than documented as retired,
and pinned upstream ids give way to our own current zen5.
50 files. Verified against the live catalog, not assumed.
The <script src="analytics.hanzo.ai/script.js" data-website-id> tag is a shim:
it talks to a read LENS instead of the one ingestion front door, so a docs site
could only ever produce pageviews — no errors, no product events, and a separate
per-site id to keep in sync. Replaced across all 12 sites with @hanzo/event
POSTing api.hanzo.ai/v1/event, which Cloud fans server-side into the web
(analytics), product (insights) and error (sentry) lenses. `product` is the only
per-site knob; Cloud resolves the tenant, so no site carries an id any more.
One implementation, not twelve: packages/analytics (@hanzo/docs-analytics)
mounts the provider, adds a pageview per client-side route change, and honors
Global Privacy Control / Do-Not-Track by disabling the client outright. The main
hub's inline Insights snippet goes too — that lens is already fed by the one
stream.
Co-authored-by: Hanzo Dev <[email protected]>
Our own docs branded a Hanzo product as an 'AI Gateway' (/docs/llm title+H1,
skills index, and ~90 files across docs/bot-docs/cloud). Renamed to 'AI API'
to match hanzo.ai/app/bot. Third-party product names are facts about someone
else's product and are NOT rewritten: 'Vercel AI Gateway' restored verbatim
(same carve-out as LiteLLM); HanzoBot's own 'Gateway' transport untouched.
Pins ghcr.io/hanzoai/static 0.4.1 -> v0.5.1: only v0.5.1 carries 79ac4f9
(serve directory index in place), so docs was the last host 301ing /docs/x/
-> /docs/x/index.html, which poisons Next's router into prefetching
.../index.html/__next._tree.txt — 8 failed requests per page load.
Purge "LLM gateway" / "unified proxy" / vs-LiteLLM framing across all
Hanzo-authored docs, marketing, and skill copy per SDK-ARCHITECTURE brand
rules. Hanzo is a full AI SDK / AI cloud, not a proxy.
- "LLM Gateway" → "AI Gateway"; hero/positioning copy → "one API for every
model". "unified proxy"/"unified gateway" removed from Hanzo copy.
- Drop "Fork of LiteLLM" / LiteLLM-as-backbone positioning. Keep genuine
third-party LiteLLM integration docs, the bot's own WebSocket Gateway
protocol, and functional config keys (litellm_params, LITELLM_MASTER_KEY)
intact.
- SDK links → canonical homes: zen-docs API table (go get
github.com/hanzo-go/sdk, pip install hanzo, npm i @hanzo/ai, cargo add
hanzo); org preset SDK entries point js-sdk→hanzo-js/sdk,
python-sdk→hanzoai/python-sdk (flagship).
Generated/synced trees (apps/spec/content/docs HIP corpus and
projects/hanzoai/bot, both gitignored) are out of this repo's committable
scope — their brand fix belongs in the source repos.
Co-authored-by: Hanzo Dev <[email protected]>
docs.hanzo.ai's AI chat now runs entirely on our own stack. The live
/api/chat route was already native (Hanzo gateway) but ungrounded — it
now retrieves from our rag-api index before answering, and cites sources:
- retrieval: rag-api /query — native index over the docs
- generation: Hanzo gateway (api.hanzo.ai/v1, zen-coder-flash)
- sources: provideLinks tool, populated from the indexed metadata
Purged Inkeep repo-wide (only our stack, one way):
- removed the dead third-party Inkeep handler (lib/inkeep -> lib/ai)
- widget credit "Powered by Inkeep AI" -> "Powered by Hanzo AI"
- create-app generator: dead `inkeep` provider -> native `hanzo`
(installs the ai/hanzo registry bundle; INKEEP_API_KEY -> HANZO_API_KEY)
- 4 legacy *-docs apps: de-Inkeeped schema, fixed dangling imports
Zero `inkeep` references remain in the codebase.
zeekay
