specification: API Commons Rate Limits specificationVersion: '0.1' schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits provider: Chroma providerId: chroma created: '2026-05-04' # Provenance stamped 2026-08-11: this artifact was written by the API Evangelist # bulk sweep dated 2026-05-04, not harvested from the provider. See roadmap#35. method: generated modified: '2026-05-05' reconciled: false tags: - AI - Vector Database - Retrieval - Serverless - Rate Limiting description: Chroma Cloud is serverless and meters Write / Storage / Query / Network rather than capping requests per second. Public documentation does not publish a numeric per-second or per-minute API rate-limit policy; tenants that exceed configured spend or per-database scaling envelopes are paused rather than throttled. The open-source Chroma server (self-hosted) has no built-in rate limiter — limits are bounded by the underlying host. notes: Chroma does not publish a numeric API rate-limit table. Tenants on Starter / Team are paused if usage exceeds limits; Enterprise can negotiate single-tenant capacity envelopes. sources: - https://www.trychroma.com/pricing - https://docs.trychroma.com/ responseCodes: throttled: 429 serverError: 500 serviceUnavailable: 503 limits: - name: Cloud usage envelope scope: tenant metric: varies limit: 'usage-bounded — Write / Storage / Query / Network are metered, not capped per-second' notes: Service pauses if monthly usage exceeds plan limits (Starter / Team) or contracted ceiling (Enterprise). - name: Self-hosted server scope: deployment metric: varies limit: 'no built-in rate limit — bounded by host CPU / RAM and any front-proxy throttling' notes: Open-source Chroma does not enforce per-client rate limits; operators add a reverse-proxy throttle if needed. policies: - name: Backoff Strategy description: On 429 / 5xx, retry with exponential backoff and jitter; honor Retry-After when present. - name: Batch Writes description: Use batched add / upsert calls rather than per-vector writes to keep the Write meter and request count efficient. - name: Pause on Quota description: Cloud tenants are paused (not throttled) when plan usage limits are exceeded; bump plan or raise the Enterprise envelope to resume. maintainers: - FN: Kin Lane email: kin@apievangelist.com