Architecture

Source

An ephemeral viewer: request, Core worker, escaped HTML

Architecture and privacy boundaries of the existing web renderer, including the branded checkout and its limits.

Document#architecture

A viewer, not a hosted document library

The existing web renderer solves a narrow problem: let a person inspect a submitted .gram file in a browser without requiring an account or storing a document collection. The brand-monochrome checkout at 254a49d is a local worktree of this same renderer repository, not an independent repository or separate product. The API accepts a multipart POST containing source bytes or text, sends it through Core, and returns render data. The /view path returns a page. Since a later GET cannot recover an earlier submission, the browser must submit the file again when it asks for another revision. This is a distinct product shape from Vault, which owns identity and storage APIs.

The separation is important for both product expectations and privacy. There is no document ID registry, account session, sync service, database, or saved sharing link in this viewer contract. The server handles submitted bytes in request memory and forgets them after responding. Request logs are documented as a fixed schema without query, header, or body values. That reduces persistence surfaces, but it does not mean the server never receives the document: the user explicitly submits bytes to the service.

Request path

A browser or command-line client sends a multipart/form-data POST to /api/v1/render, /api/v1/search, or /view. A field chooses one input representation (file or source) and optional selectors such as revision, at, history, history_offset, or html. The service validates field cardinality and size, then invokes Core in a fresh worker process. The worker receives document input on stdin and returns structured results on stdout. Time, memory and concurrency limits bound the work; failures are explicit errors such as E_LIMIT, E_STALE, E_BUSY, or E_TIMEOUT.

Core history and replay supply the selected source and projections. The HTML renderer escapes document values, restricts links to safe schemes, and does not emit document-provided scripts, styles, or images. Unsupported blocks are retained as source with an unsupported-content indication. The browser keeps a selected file in tab memory and posts it again for state changes; the documented implementation avoids browser Storage APIs, service workers, and putting document text in the URL. A content security policy and no-store responses further constrain page behavior.

Example: render one selected historical state

For a local file named plan.gram, a client can submit file=@plan.gram and revision=2 to /api/v1/render; to ask for an HTML fragment, include html=1. A caller should handle a 422 Core history error, a 413 size cap, and temporary 503 worker saturation rather than claiming every source produces a complete view. The renderer may return summary metadata for old states while some detailed change projections are bounded by a source-byte budget. A rendered view is a projection, not a rewrite of the original document.

The branded worktree is not a separate service

The local brand-monochrome git worktree is based on thegram-web-render at earlier commit 254a49d. Its remote is the same renderer repository; it is not a separate product repository or independent service. The difference is a local snapshot, not evidence of another identity store, deployment, or retention model.

Tradeoffs and limits

A stateless renderer avoids account friction and server-side document libraries, but it requires sending the bytes on every interaction and does not create a durable share URL. HTML output is safe by default and may omit unsupported rich content or show raw source; this is not a full fidelity promise for every format construct. Resource caps protect service availability but mean a large file or busy worker pool can fail. Public access and ephemeral handling do not equal confidentiality guarantees beyond the described processing and logging boundary. A user handling sensitive data should decide whether the service is appropriate before submitting it.

Read docs/api.md for the exact multipart contract and error codes, README.md for browser behavior and fidelity, and deploy/README.md for deployment controls recorded by that checkout. Do not infer current deployment status from source configuration alone.

Sources

Implementation claims use both renderer checkouts' README, API and deployment documentation.

Source

Source
Renderer README and browser flow

Internal source reviewed at revision 971a0474fb4b97c3d23d6687be80664461f43a2f (main: README.md); not publicly available.

Source
Renderer HTTP API contract

Internal source reviewed at revision 971a0474fb4b97c3d23d6687be80664461f43a2f (main: docs/api.md); not publicly available.

Source
Renderer deployment configuration documentation

Internal source reviewed at revision 971a0474fb4b97c3d23d6687be80664461f43a2f (main: deploy/README.md); not publicly available.

Source
Branded renderer README

Internal source reviewed at revision 254a49d8adb6ccb98675a0511283f511914394c9 on branch brand-monochrome (README.md); not publicly available.

Source
Branded renderer API contract

Internal source reviewed at revision 254a49d8adb6ccb98675a0511283f511914394c9 on branch brand-monochrome (docs/api.md); not publicly available.

Source
Branded renderer deployment documentation

Internal source reviewed at revision 254a49d8adb6ccb98675a0511283f511914394c9 on branch brand-monochrome (deploy/README.md); not publicly available.

Knowledge

Knowledge
Renderer processes submitted documents ephemerally

The request path uses worker processes and in-memory document processing without a persistent document store in the documented contract.

→ Supports: A bounded worker turns Core output into HTML
Knowledge
A bounded worker turns Core output into HTML

Fresh workers, resource limits, escaping and restricted browser behavior constrain untrusted document rendering.

→ Supports: The branded checkout is not proof of a separate service
Knowledge
The branded checkout is not proof of a separate service

Similar API and deployment docs describe a checkout distinction; operational independence requires separate evidence.

Knowledge
A stateless viewer still receives document bytes

No storage does not mean no transmission; a user must decide whether the chosen document is appropriate to submit.

→ Supports: Renderer processes submitted documents ephemerally

An ephemeral viewer: request, Core worker, escaped HTML

r4 · · Unknown author

Architecture and privacy boundaries of the existing web renderer, including the branded checkout and its limits.

0/0Tasks done
0Needs attention

Source6

Renderer README and browser flow
Renderer HTTP API contract
Renderer deployment configuration documentation
Branded renderer README
Branded renderer API contract
Branded renderer deployment documentation

Knowledge4

Renderer processes submitted documents ephemerally
A bounded worker turns Core output into HTML
The branded checkout is not proof of a separate service
A stateless viewer still receives document bytes

Source and build provenance

Renderer API: thegram-web-render/1. Core source: git+https://github.com/clab-one/thegram-format.git?rev=625edecee877ddf5f6c303cb478c64fa4cbca651#625edecee877ddf5f6c303cb478c64fa4cbca651.

The download file hash covers exact .gram bytes; the renderer Source sheet reports the source used for its selected revision.

This page was generated from the canonical viewer-architecture.gram source.

Download file SHA-256
d9b30d7381a0d2731ffaccd9fdbd819f237ef774f42da2c81ff61a0393198d16
Renderer
Existing local Thegram renderer (local Thegram renderer); rendered at build time.
Renderer document
blog-viewer-architecture
Selected revision
r4

Revision navigation and historical search are not offered on this static publication. The displayed revision is the one rendered from this exact source during the build.