Fonts & Resource Limits¶

These are the numbers that decide how much a self-hosted instance can take.

CPU limits¶

docker-compose.yml sets a CPU quota on both containers by default:

Container	Setting	Equivalent
`frontend`	`cpu_quota: 50000`, `cpu_period: 100000`	half a core
`backend`	`cpu_quota: 80000`, `cpu_period: 100000`	0.8 of a core

To remove the limit, delete or comment out those three lines (cpu_count / cpu_quota / cpu_period). To change how strict it is, tune cpu_quota — it is "how many microseconds of CPU per cpu_period microseconds".

Memory limits are mem_limit / memswap_limit: 800 MB for the frontend and 1500 MB for the backend. Rendering large sheets is memory-hungry — if you see out-of-memory errors, start here.

Concurrency: two renders at a time¶

The backend has a concurrency gate:

MAX_CONCURRENT_EXECUTIONS = 2

At most two render jobs run at once; the rest queue. This is sized for a modest machine — rendering is CPU-bound, so running more at once mostly just slows everything down. On a powerful machine you can raise it, but raise cpu_quota too or they will still be fighting over 0.8 of a core.

Queue: returns 503 when full¶

MAX_ACTIVE_TASKS = 8

When "queued + processing" reaches 8, new submissions are rejected outright with a 503, which the frontend turns into a wait countdown. The backlog is bounded.

Task expiry¶

TTL = 30 minutes

Expiry is measured from the task's last update. Submission, status, and result requests clean up expired tasks and their response files. Download within that window; expired task IDs return 404. Temporary files also have scheduled cleanup, but disk usage still needs monitoring.

Overload protection¶

The backend samples system CPU usage and returns 429 for new requests above the threshold:

CPU_USAGE_LIMIT=90   # default 90 (percent); 100 effectively disables it

The E2E tests and the desktop build set it to 100 precisely so this guard stays out of the way.

Text length and page count¶

There is no fixed ten-page truncation or domain-specific text-length restriction. CPU protection, queue capacity, paper pixel validation, memory, and proxy timeouts still constrain long documents. Read backend/app.py for the actual checks.

Rate limiting¶

Endpoints are rate-limited per client IP, defaulting to:

1000 per 5 minute

Individual routes have their own, more specific limits (polling and fetching results are far more generous than submitting a job). When you are throttled you get a 429.

Fonts¶

Fonts baked into the Docker image come from the repository's ttf_files/, copied to /app/font_assets at build time.
Your own fonts go in ttf_files/ — see Docker Deployment.
Both appear in the font dropdown.

Check font licences yourself

The code is MIT; the fonts are not. Fonts bundled in the repository and fonts users upload each carry their own licence. Before serving them to others, confirm you are allowed to redistribute them.