Prakasa (Sanskrit: प्रकाश) — The decentralized "Light" of intelligence.
Prakasa is a decentralized, privacy-preserving P2P GPU inference middleware built on top of the open-source Parallax engine. It leverages the Nostr protocol for resilient, censorship-resistant orchestration and integrates the RIM economic system for trustless, real-time settlements and reputation.
Prakasa is a high-performance, privacy-centric P2P GPU inference middleware that transforms idle GPU resources into a unified, resilient intelligence layer. By evolving the Parallax orchestration engine and integrating modern decentralized primitives, Prakasa enables secure, incentivized model deployment across heterogeneous devices.
-
Nostr-Powered Orchestration: Uses the Nostr protocol as a decentralized communication bus.
- Resilience: No single point of failure for node discovery and task distribution.
- Privacy: Uses NIP-44/59 (Gift Wrap) for encrypted metadata and secure client-provider signaling.
- Interoperability: Integrates with the broader Nostr ecosystem for monitoring and management.
-
RIM Economic System: Robust Incentive Mechanism adapted for P2P compute markets.
- Atomic Settlements: Real-time micropayments per inference.
- Market Pricing: Dynamic supply-demand pricing to fairly compensate providers.
- Sybil Resistance: Staking and reputation to ensure high-quality compute.
-
Privacy & Hardened Security:
- Encrypted Payloads: End-to-end encryption of model weights and I/O.
- Isolated Execution: Stronger containerization for multi-tenant GPU workloads.
- Verifiable Computation (Roadmap): Zero-knowledge proofs for verifiable results.
-
Evolution of Parallax: Built on Parallax’s core, extended for global P2P scale.
- Stateless discovery replacing central APIs with Nostr relays.
- Modular settlement integrating the RIM protocol.
Prakasa is organized as a middleware stack bridging raw hardware and AI applications:
- Infrastructure Layer: Global P2P GPU nodes running the Prakasa-Parallax runtime.
- Communication Bus (Nostr): Task relaying, heartbeat monitoring, encrypted handshakes, and discovery.
- Settlement Layer (RIM): Manages instant value transfers ("Spanda") and on-chain/off-chain reconciliations.
- Application Layer: SDKs and APIs for developers to consume compute-as-a-service with privacy and incentives built-in.
- Installation and local dev instructions: see docs/user_guide/install.md
- Quick start: see docs/user_guide/quick_start.md
We welcome contributions! See the Contributing Guide for how to get started. If you'd like Prakasa to integrate a specific settlement backend, relay implementation, or privacy primitive, please open an issue.
- Prakasa is designed as an evolution and distribution of the Parallax project; this repository contains the foundational Parallax engine and components. Prakasa-specific modules (Nostr integration, RIM connectors, settlement adapters) are maintained in dedicated submodules and extensions.
The project also includes a catalog of supported models and integrations. See the table below for currently supported model families and collections.
| Provider | HuggingFace Collection | Blog | Description | |
|---|---|---|---|---|
| DeepSeek | Deepseek | DeepSeek-V3.2 DeepSeek-R1 |
Deep Seek AI Launches Revolutionary Language Model | Deep Seek AI is proud to announce the launch of our latest language model, setting new standards in natural language processing and understanding. This breakthrough represents a significant step forward in AI technology, offering unprecedented capabilities in text generation, comprehension, and analysis. |
| MiniMax-M2 | MiniMax AI | MiniMax-M2 | MiniMax M2 & Agent: Ingenious in Simplicity | MiniMax-M2 is a compact, fast, and cost-effective MoE model (230B parameters, 10B active) built for advanced coding and agentic workflows. It offers state-of-the-art intelligence and coding abilities, delivering efficient, reliable tool use and strong multi-step reasoning for developers and agents, with high throughput and low latency for easy deployment. |
| GLM | Z AI | GLM-4.7 GLM-4.6 |
GLM-4.7: Advancing the Coding Capability | "GLM" is an advanced large language model series from Z AI, including GLM-4.6 and GLM-4.7. These models feature long-context support, strong coding and reasoning performance, enhanced tool-use and agent integration, and competitive results across leading open-source benchmarks. |
| Kimi-K2 | Moonshot AI | Kimi-K2 | Kimi K2: Open Agentic Intelligence | "Kimi-K2" is Moonshot AI's Kimi-K2 model family, including Kimi-K2-Base, Kimi-K2-Instruct and Kimi-K2-Thinking. Kimi K2 Thinking is a state-of-the-art open-source agentic model designed for deep, step-by-step reasoning and dynamic tool use. It features native INT4 quantization and a 256k context window for fast, memory-efficient inference. Uniquely stable in long-horizon tasks, Kimi K2 enables reliable autonomous workflows with consistent performance across hundreds of tool calls. |
| Qwen | Qwen | Qwen3-Next Qwen3 Qwen2.5 |
Qwen3-Next: Towards Ultimate Training & Inference Efficiency | The Qwen series is a family of large language models developed by Alibaba's Qwen team. It includes multiple generations such as Qwen2.5, Qwen3, and Qwen3-Next, which improve upon model architecture, efficiency, and capabilities. The models are available in various sizes and instruction-tuned versions, with support for cutting-edge features like long context and quantization. Suitable for a wide range of language tasks and open-source use cases. |
| gpt-oss | OpenAI | gpt-oss gpt-oss-safeguard |
Introducing gpt-oss-safeguard | gpt-oss are OpenAI’s open-weight GPT models (20B & 120B). The gpt-oss-safeguard variants are reasoning-based safety classification models: developers provide their own policy at inference, and the model uses chain-of-thought to classify content and explain its reasoning. This allows flexible, policy-driven moderation in complex or evolving domains, with open weights under Apache 2.0. |
| Meta Llama 3 | Meta | Meta Llama 3 Llama 3.1 Llama 3.2 Llama 3.3 |
Introducing Meta Llama 3: The most capable openly available LLM to date | "Meta Llama 3" is Meta's third-generation Llama model, available in sizes such as 8B and 70B parameters. Includes instruction-tuned and quantized (e.g., FP8) variants. |