pllmReimplement independently
Research papers are specification and provenance inputs. Implementations are built in PLLM's Rust and Python stack rather than imported as runtime dependencies.
Research for high-performance private LLM inference.
PLLM is a high-performance private LLM multi-party inference runtime and autonomous research harness. We independently reimplement useful methods, compose compatible components, and benchmark complete plans to determine what improves the system.
A runtime and a research system.
The runtime keeps plaintext, model state, masks, and decoding inside the client boundary while separate services perform prepared masked computation. The research harness turns pinned papers and hypotheses into reusable components, immutable plans, assurance records, and matched benchmark evidence.
Compose by capability
Like-for-like cache, numeric, protocol, preparation, kernel, and placement components live together behind stable typed contracts.
Compare complete plans
Correctness and privacy remain hard gates. Latency, throughput, traffic, memory, preparation, quality, energy, and cost remain separate measurements.
Three research records.
Each record serves a different reader. The whitepaper explains why PLLM exists, the technical paper defines the implemented system and its limits, and the paper catalog tracks the methods that may improve it.
01
PLLM original research
A concise, business-friendly whitepaper on private inference, the multi-party boundary, and the autonomous research loop.
Read the whitepaper → 02PLLM research paper
The current protocol, compiler and component architecture, security limits, implementation status, and retained evidence.
Read the technical paper → 03Papers
One PLLM-focused page per tracked source, with its contribution, boundary, status, original source, and implementation links where they exist.
Explore the catalog → 04Reimplementation backlog
The ordered work needed to turn tracked methods into validated components and comparable plans.
See what comes next →