ElizaOS RL training pipeline with benchmarking and model publishing support

STZ Foundry: the evolution of slice-tournament-zoo into a standalone BYO-LLM foundry — a contract-bounded slice pipeline that implements each slice adversarially via an N-specimen tournament with frozen sealed tests, GRPO-style selection, layered anti-rew

STZ: a contract-bounded slice pipeline that implements each slice adversarially via an N-specimen tournament with frozen sealed tests, GRPO-style selection, layered anti-reward-hacking, a replayable markdown audit trail, and (0.9.0) a bounded harness-leve

No description provided.

Install AI research engineering skills to your coding agents (Claude Code, OpenCode, Cursor, Gemini CLI, Hermes Agent, and more)

Prototype state/action/reward environment primitives for Ignition Agent Trainer.

Domain-neutral visual primitive engine for the web. Declarative SVG diagram/math primitives, draggable handles, a math↔pixel coordinate system, a motion core + energy effect defs, theming, pure sim cores, and a small kit of neutral glyph primitives. React

Codebase intelligence, recursive self-improvement pipeline, and daemon system for HoloScript

PostTrain: run and track post-training (data, RL environments, SFT, preference tuning, RL, evals, deploy) from the terminal or the browser.

Command-line access to the Belvedir platform: projects, keys, traces, training, review, benchmarks, environments, organizations, routing, and billing.

Internal native bindings - import from @mlx-node/lm or @mlx-node/trl instead

Experimental RL-inspired utilities for Ignition Agent Trainer.

Fine-tune a small LLM on your website's own pages (local LoRA/QLoRA via Unsloth on NVIDIA or MLX on Apple Silicon) and embed it as a WebGPU chat window that serves the model tier matching each visitor's hardware. Generic — works with any static site folde

Internal component of Sutra SDK — applications should install sutra-sdk. Learning substrate contracts: turn-trajectory schema (B9 metadata + C0 training fields), parse boundary for corpus/gym/critic pipelines.

Open-source CLI coding agent with 22-layer persistent memory. Free with Ollama. Drop-in Claude Code alternative. Works with any model.

Research-lab layer for OpenCode: six personas, 17 curated skills, and a typed claim-level lab record with provenance that survives the chat.

Synthetic data generation for AI agent training