Tools Directory

Open-source CLI that runs static analysis on prompt templates, flags tokenmaxxing patterns — redundant instructions, scaffolding residue, over-specified format directives — and generates a compression diff with quality-preservation annotations. Git integration surfaces token cost deltas on every commit.

Token observability platform with per-call attribution, prompt template tracking, and trend dashboards. Integrates with OpenTelemetry. Configurable soft and hard budget guardrails via API or YAML. Supports all major inference providers and self-hosted deployments.

Model routing library for Python and TypeScript. Classifies incoming requests against a configurable complexity taxonomy and dispatches to the appropriate model tier. Built-in per-route cost tracking and latency telemetry. Ships with pre-trained classifiers for common NLP task types.

Token budget enforcement middleware for agentic frameworks. Wraps any orchestration layer with configurable per-run, per-step, and per-session token ceilings. Supports graceful termination with state checkpointing. Compatible with LangChain, LlamaIndex, and bare API loops.

RAG precision toolkit providing chunk quality scoring, dynamic similarity threshold tuning, and re-ranker integration. Benchmarks retrieval precision against annotated ground truth sets and recommends chunk size and overlap parameters to minimize context stuffing without degrading recall.

Stateful context compression library for multi-step agentic workflows. Implements working memory and episodic memory separation with pluggable summarization backends. Tracks context window utilization and compression ratios per agent run and exposes a dashboard for tail-behavior analysis.