---

name: palantir-rate-limits description: >- Analyze and design bounded Foundry API concurrency, backoff, pagination, and overload behavior from current documented limits. Use when preventing or handling HTTP 429 or 503 responses. Trigger with "Palantir rate limit". allowed-tools: Read,Glob,Grep,Write,Edit version: 2.0.0 argument-hint: "[client-or-endpoint]" model: inherit effort: high author: Jeremy Longshore jeremy@intentsolutions.io license: MIT compatibility: Requires current Palantir Foundry documentation and approved access for any live resource, permission, data, build, application, or deployment change tags: [saas, palantir, foundry, rate-limits, reliability]

Palantir API Limit and Concurrency Control

Overview

Protect Foundry and the caller from retry storms. Budget requests across the acting user, bound concurrent work, honor server responses, and distinguish retryable throttling from authorization, validation, or permanent resource errors.

Prerequisites

• Identify the acting user or service user, applications sharing it, endpoints, workload peaks, page/batch shape, and service objective.
• Capture response status, request identifiers, timestamps, server timing guidance, retry count, and current concurrency.
• Read references/official-docs.md immediately before setting numeric budgets because effective and endpoint-specific limits may change.
• Prepare load tests in an approved non-production scope.

Current Contract

• The current general API documentation lists global per-user rate and concurrency limits across Foundry API endpoints.
• Individual endpoints may impose stricter limits and return 429 or 503.
• Palantir recommends exponential backoff for throttled requests.
• Limits can vary in practice, and disruptive throttling may require Palantir Support rather than client-side limit evasion.

Authentication

Attribute the request budget to the actual OAuth principal. Do not distribute calls across extra tokens, users, or service users to evade a limit. Keep token material out of metrics and retry logs.

Instructions

1. Inventory every worker and application using the same principal and estimate peak request rate plus in-flight concurrency.

2. Set a conservative shared concurrency budget and bounded queue with deadlines and cancellation.

3. Retry only documented transient statuses, honoring server guidance and using capped exponential backoff with jitter.

4. Use pagination, selective properties, batching supported by the endpoint, idempotency controls, and workload coalescing to reduce demand.

5. Load-test below the approved ceiling, simulate 429 and 503, and verify recovery without duplicate writes or unbounded backlog.

Tool Discipline

• Use Glob to locate candidate repositories, manifests, configurations, and evidence without widening scope.
• Use Grep to find relevant identifiers, declarations, permissions, errors, and stale claims.
• Use Read to inspect the smallest required files and authoritative evidence.
• Use Write only for a new approved local draft, test, manifest, or evidence artifact.
• Use Edit only for a bounded approved change whose rollback is known.
• Do not use file tools as a substitute for authenticated Foundry operations or owner approval.

Approval Boundaries

Application owners approve budgets and degradation; platform owners approve load tests; data owners approve any write retry or batch behavior. Creating principals to bypass limits is prohibited.

Output

A principal-level request inventory, endpoint budgets, concurrency and queue design, retry matrix, idempotency rules, test evidence, alert thresholds, and support escalation criteria.

Error Handling

Condition	Response
A non-retryable 4xx is retried	Stop and route it to validation, authentication, authorization, or resource triage.
Queue age exceeds the service objective	Reject or degrade work according to policy instead of accumulating unbounded backlog.
Retries duplicate a write	Disable automatic retry until idempotency or reconciliation is proven.
Throttling persists below the expected budget	Check shared-principal traffic and endpoint limits, preserve request IDs, and contact Palantir Support.

Examples

Example 1

Coordinate several workers under one client-credentials service user with a shared semaphore, bounded queue, capped jittered backoff, and metrics for 429, 503, in-flight requests, and queue age.

Example 2

Make an object reader cheaper and safer by selecting required properties, paging deterministically, cancelling expired work, and lowering concurrency before increasing retries.

Validation

• Request and concurrency budgets are enforced across all users of the principal.
• Retry attempts and elapsed time have hard bounds.
• Write operations are idempotent or reconciled before retry.
• Load tests recover from throttling without a storm or duplicate side effects.
• Alerts identify principal, endpoint, queue age, and request IDs without exposing tokens.

Resources

• Official documentation and contract notes
• Re-check the dated contract before any live operation.
• Treat unresolved or changed vendor behavior as a stop condition.