SynapsesOS
Reference

Rate Limits

Rate Limits

Token-bucket rate limiting prevents runaway agents from overwhelming Synapses. Each limit is per-session.

{
  "rate_limits": {
    "write_ops_per_minute": 30,
    "expensive_reads_per_minute": 20,
    "cross_project_per_minute": 60
  }
}

Fields

FieldTypeDefaultDescription
write_ops_per_minuteint30Limit for write operations: memory(action="save"), annotate(action="add")
expensive_reads_per_minuteint20Limit for memory(action="search") calls (involve embedding + search)
cross_project_per_minuteint60Limit for cross-project federation queries

How It Works

Rate limits use a token-bucket algorithm:

  • Each bucket starts full (at the per-minute limit)
  • Each operation consumes one token
  • Tokens refill at a rate of limit / 60 per second
  • When the bucket is empty, the operation returns a rate limit error

When to Adjust

  • High-frequency agents: If an agent makes many memory(action="search") calls, increase expensive_reads_per_minute to 40
  • Multi-agent environments: Lower write_ops_per_minute to 15 per agent to prevent contention
  • Heavy federation use: Increase cross_project_per_minute to 120 for monorepo workflows