Skip to content

limits.md

Latest commit

 

History

History
96 lines (71 loc) · 3.48 KB

limits.md

File metadata and controls

96 lines (71 loc) · 3.48 KB
title Limits and quotas
description What's bounded on the platform and by how much.
order 6

Limits and quotas

Hard and soft limits enforced by the platform. Many are configurable in the Helm chart; defaults are listed here.

File and message limits

Limit Value Notes
Per-file upload size 20 MB Hard limit. Larger files require a custom MCP server that chunks content.
Per-message file attachments 10 Soft limit; configurable in chat.maxAttachmentsPerMessage.
Per-message body length 64 KB Limit on Markdown body; longer messages should attach as files.

Agent and workload limits

Limit Default Notes
Agents per organization Unlimited Subject to resource quota.
MCP servers per agent 32 Soft limit; can be raised.
Skills per agent 256 Soft limit.
Hooks per agent 32 Soft limit.
ENV variables per resource 64 Practical Kubernetes ceiling.
ENV total size per container 32 KB Kubernetes hard limit.
Volumes per agent (including sub-resources) 16 Soft limit.
Volume size 1 TB Hard limit per volume. Configurable in agents.maxVolumeSizeGb.
Concurrent workloads per runner Capped by runner resources Not enforced by the platform.
Workload idle timeout 5m default, 1m–24h range Configurable per-agent.

API rate limits

Limit Default Notes
Per-token Gateway requests 600 / minute Configurable in gateway.rateLimit.
Per-IP Gateway requests 3000 / minute Configurable in gateway.rateLimit.
LLMProxy per-token requests Provider-dependent LLM Proxy passes through provider rate limits; configure per-provider concurrency.

Authorization

Limit Value
Tuples per check OpenFGA's per-store limit (very high)
Relations evaluated per ListObjects Capped by OpenFGA's resolution depth (default 25)
Max tuples per Write 100

Tracing

Limit Default Notes
Per-span size 64 KB Hard limit. Larger spans are rejected.
Spans per run 10000 Soft limit; runs exceeding this are truncated.
Run retention 14 days Configurable in tracing.retentionDays.

OpenZiti

Limit Notes
Identities per controller OpenZiti-defined; depends on controller deployment.
Concurrent flows per identity OpenZiti-defined; depends on routers.

Devices

Limit Value
Devices per user Unlimited
JWT enrollment validity 7 days (configurable per IdP setup)

API tokens

Limit Value
API tokens per user Unlimited
Token TTL Optional. Unlimited by default.

Apps

Limit Default Notes
Apps published per organization Unlimited
Installations per app Unlimited
Audit log per installation 1000 events Ring buffer — oldest dropped.

Adjusting limits

Soft limits configurable per service live in that service's chart values — bootstrap renders them inline in stacks/platform/main.tf (look for <service>_values). For an off-bootstrap install, pull the chart with helm pull oci://ghcr.io/agynio/charts/<service> and read its values.yaml.

Hard limits (file size, ENV total size) are platform-design choices and cannot be overridden — they reflect Kubernetes ceilings or platform-side guarantees we don't want to weaken.

Related