#structured-output

Schema-validated JSON blocks for AI-generated reports — Markdown source, safe runtime, zero AI-generated HTML/CSS/JS. Works with Claude Code, Codex, any agent.

OpenAI-compatible HTTP API for Needle 2, Cactus Compute's 45M-parameter tool-calling model. Drop-in for any OpenAI client.

Model-agnostic substrate for running small local LLMs well — code agents, chat, image/video, speech. Small capable main model + tiny fast worker.

Turn natural-language search queries into structured JSON filters using the Claude API. Production-grade intent extraction pattern.

Experimental opt-in Jev skills and config for Claude Code — the companion to jev-codex-tools.

Recover a structured result from an agent session: its structured_output, then the StructuredOutput inputs the CLI rejected, then the JSON in its result text and its assistant texts.

Keep It Real — a schema-level deferral channel that makes LLM brevity safe. Code, data, and pre-registered protocols: a 400-conversation replication sweep, a placebo control, and a blinded human study (n=46).

Deterministic repair and validation for LLM JSON output. 8% of malformed responses parse raw; 96% after the repair ladder.

Typed, state-first agent pipelines for .NET & TS, built on Microsoft Agent Framework.

Fetch a website and use an LLM to extract structured profile data against a user-supplied schema. Pluggable Gemini / OpenAI / Anthropic backends. Lead enrichment without Clearbit.