Prime Intellect

Prime Intellect is an open, decentralized stack for developing and improving AI agents through reinforcement learning. This API Evangelist catalog entry profiles the company's full developer surface: GPU compute marketplace, hosted RL post-training ("Lab"), secure remote sandboxes, OpenAI-compatible inference, evaluations against the Environments Hub, and account/team platform APIs.

Provider

• Website: https://www.primeintellect.ai
• Docs: https://docs.primeintellect.ai
• App / Dashboard: https://app.primeintellect.ai
• API Reference: https://docs.primeintellect.ai/api-reference/introduction
• OpenAPI: https://api.primeintellect.ai/openapi.json
• GitHub org: https://github.com/PrimeIntellect-ai

APIs

API	Base URL	OpenAPI
Prime Intellect Compute API	https://api.primeintellect.ai	openapi/prime-intellect-compute-api-openapi.yml
Prime Intellect Sandbox API	https://api.primeintellect.ai	openapi/prime-intellect-sandbox-api-openapi.yml
Prime Intellect Training API	https://api.primeintellect.ai	openapi/prime-intellect-training-api-openapi.yml
Prime Intellect Evaluations API	https://api.primeintellect.ai	openapi/prime-intellect-evaluations-api-openapi.yml
Prime Intellect Inference API	https://api.pinference.ai/api/v1	openapi/prime-intellect-inference-api-openapi.yml
Prime Intellect Platform API	https://api.primeintellect.ai	openapi/prime-intellect-platform-api-openapi.yml

All control-plane surfaces share the same bearer-token API key (Authorization: Bearer $PRIME_API_KEY). The inference plane runs on a separate hostname (api.pinference.ai) and uses the same key.

Artifacts

• apis.yml — APIs.json index for this provider
• openapi/ — six OpenAPI 3.1 specs split by product surface, derived from the upstream api.primeintellect.ai/openapi.json (Compute, Sandbox, Training, Evaluations, Platform) plus an authored Inference spec for api.pinference.ai
• capabilities/ — Naftiko 1.0.0-alpha2 capabilities (one per business surface) with REST and MCP adapters
• json-schema/ — Core entity schemas (Pod, Sandbox, Training Run, Evaluation)
• json-ld/ — Linked-data context covering Pod, Sandbox, TrainingRun, Evaluation, Disk, Wallet, Usage
• plans/ — API Commons Plans 0.1 for compute, training, and inference pricing
• rate-limits/ — API Commons Rate Limits 0.1 (best-effort; not all limits are publicly documented)
• finops/ — FOCUS-aligned FinOps definition spanning GPU-hours and token meters

CLI and SDK

• pip install prime — official CLI + Python SDK (prime pods, prime sandbox, prime env, prime train, prime eval, prime teams)
• pip install prime-sandboxes — lightweight (~50KB) sandbox-only client
• Source: https://github.com/PrimeIntellect-ai/prime

Open Source

• verifiers — library for RL environments and evaluations
• prime-rl — agentic RL training at scale
• renderers — programmable chat templates for LLM training and inference
• research-environments — environments by the Prime Intellect research team
• community-environments — community-contributed environments