Anthropic June 15 per-subscription cap:
your agents will hit the wall.
Starting June 15, Anthropic enforces a hard programmatic cap per subscription plan. One Anthropic account means one cap. When your agent swarm hits it, production stops.
What the cap actually means for your agents
Each Anthropic subscription has its own daily/monthly request limit. All your agents share that single ceiling.
When the cap hits, requests return 429 errors. No warning, no slowdown. Your agent loop dies mid-task.
You cannot buy more capacity on the fly. You wait for the billing cycle. Every minute of downtime is real cost.
How RelayPlane solves the June 15 cap
Add multiple Anthropic API keys or accounts. RelayPlane routes across them automatically. Each account stays under its own cap. Your agents never see a 429.
1. Credential pool spreads load across accounts
Configure a primary Anthropic key plus fallback keys in your relayplane.config.yml. The proxy round-robins requests across the pool. Each key's usage stays proportionally distributed, so no single account approaches its cap while others sit idle.
2. Quota-aware router fails over before the cap hits
RelayPlane tracks per-credential usage in real time against configurable thresholds. When a key approaches its limit, the router automatically shifts traffic to the next credential in the pool. No 429s reach your agent code.
3. 30-second install, one environment variable change
RelayPlane is a local npm proxy. Your agents already call the Anthropic API. You change one environment variable to point at localhost. No SDK migration, no code rewrites, no new auth flows.
Get running before June 15
1. Install and init
# Install globally
npm install -g @relayplane/proxy
# Initialize config
npx relayplane init
# Start the proxy
relayplane start2. Add your credential pool
# relayplane.config.yml
providers:
anthropic:
credentials:
- key: ${ANTHROPIC_KEY_1} # primary
weight: 1
- key: ${ANTHROPIC_KEY_2} # fallback
weight: 1
- key: ${ANTHROPIC_KEY_3} # extra capacity
weight: 1
quota:
soft_limit: 0.85 # fail over at 85% cap3. Point your agents at the proxy
# One env var change. Nothing else in your agent code changes.
export ANTHROPIC_BASE_URL=http://localhost:4100
# Your agent still calls anthropic.messages.create() normally.
# RelayPlane intercepts, routes across your credential pool,
# and returns the response. Each key stays under its cap.Full docs at relayplane.com/docs/quickstart
Why no other npm-native gateway has this story as cleanly
Primary + fallback key chain per provider. Not a workaround, a first-class feature.
Tracks usage per credential in real time. Fails over before the 429 hits your agent.
Works with any agent using the Anthropic SDK or OpenAI-compatible clients.
Keys never leave your network. No cloud dependency, no new auth surface.
One environment variable. Your agent loops keep running unchanged.
No vendor lock-in. Audit it, fork it, run it forever.