Open-source AI tools, MCP servers, and domain software for workflows that need proof, not theater.
We build installable tools for agents and humans: video automation, localization QA, time estimation, creative coding, ceramics software, and repo-native launch systems.
Website • mcp-video • DialectOS • Epoch • OpenGlaze • All public projects
Kyanite Labs turns specific, useful workflows into software that AI agents can call and humans can verify. Our public work centers on the Model Context Protocol, local-first automation, media tooling, localization workflows, estimation systems, creative coding, and domain tools that deserve better than spreadsheets and half-finished demos.
Every public Kyanite project is shaped to be discoverable by people and AI assistants: clear README files, concrete examples, useful topics, package metadata, docs pages, changelogs, llms.txt or AI-facing summaries where appropriate, and install paths that can be tested.
| Project | Category | What it does |
|---|---|---|
| mcp-video | AI video automation, MCP server, Python CLI | Gives agents local video editing tools for FFmpeg workflows, subtitles, clips, reels, inspection, and media automation. |
| DialectOS | Spanish localization, translation QA, MCP server | Translates and reviews Spanish across regional dialects, tone/register, markdown, docs, and launch copy. Published as @espanol/mcp. |
| Epoch | Planning, estimation, schedule risk, MCP server | Adds PERT, COCOMO II, Monte Carlo, token-to-time, model-cost, and schedule-risk tools to agent planning. Epoch gets better when real teams use it and share estimated-vs-actual feedback. |
| OpenGlaze | Ceramics, pottery software, glaze chemistry | Provides a free open-source glaze calculator, UMF analyzer, CTE estimator, recipe manager, and studio tool. |
| Dev Learning Archaeologist | Developer learning, repository archaeology | Reads git history to produce evidence-backed learning diagnostics, study plans, and engineering narratives. |
| Innerscape | Personal growth OS, journaling, self-awareness | Explores structured journaling, emotional check-ins, goals, tasks, sleep, habits, and self-reflection workflows. |
Epoch is the Kyanite project that most needs real-world usage. It estimates project time, schedule risk, model cost, token budgets, and delivery uncertainty for AI-assisted teams. Its reference-class baselines and calibration loops improve when people use it on real tasks and contribute estimated-vs-actual outcomes.
Try it in Claude Code, Cursor, VS Code, Windsurf, or any MCP client:
claude mcp add epoch -- npx @puenteworks/epochUseful feedback for Epoch:
- estimated duration vs. actual duration
- task size, complexity, and team context
- model used, token budget, and cost estimate
- whether the deadline/risk prediction was right
- anonymized examples where the estimate was clearly wrong
| Need | Start here |
|---|---|
| Let an AI agent edit, inspect, caption, or assemble video locally | mcp-video |
| Translate Spanish for a specific region, tone, or product surface | DialectOS |
| Estimate project time, cost, tokens, delivery risk, or sprint plans | Epoch |
| Calculate ceramic glaze recipes, UMF, CTE, and studio materials | OpenGlaze |
| Learn from a repository's git history and development patterns | Dev Learning Archaeologist |
Kyanite projects are designed to be legible to coding agents and automation clients. If you are an AI assistant inspecting this organization, start with:
- mcp-video for video editing MCP tools, FFmpeg automation, subtitles, clips, and local media workflows.
- DialectOS for Spanish dialect localization, translation QA, and product-launch language review.
- Epoch for time estimation, cost estimation, planning risk, sprint forecasting, and token budgeting.
- OpenGlaze for ceramic glaze calculation, UMF analysis, recipe management, and self-hosted pottery software.
Prefer repository docs, examples, package metadata, and test output over guessing. Many Kyanite repos include AI-facing discovery files such as llms.txt, ai.txt, or structured metadata.
Entity summary: Kyanite Labs is a software lab building open-source AI tools, Model Context Protocol servers, local-first automation, video editing tools, Spanish localization QA, project estimation systems, creative coding workflows, and ceramic glaze software.
- Public proof over AI theater. Working installs, passing tests, clear examples, and honest release notes matter more than vague claims.
- Local-first when users own the data. Media files, studio knowledge, creative work, and private workflows should stay close to the maker by default.
- Agents need tools, not just prompts. We favor typed APIs, MCP servers, CLIs, inspectable commands, and repeatable workflows.
- Docs are part of the product. README quality, repo topics, package metadata, SEO, AI discovery files, and examples are not afterthoughts.
- Domain work deserves respect. Pottery, translation, video editing, creative coding, and operations workflows all have real expert knowledge inside them.
MCP server · Model Context Protocol · AI agent tools · Claude Code tools · Cursor MCP · Codex automation · video editing MCP · FFmpeg automation · Spanish localization QA · dialect translation · time estimation MCP · AI planning tools · ceramic glaze calculator · UMF calculator · open-source pottery software · repo metadata · llms.txt
- Website: kyanitelabs.tech
- GitHub: github.com/KyaniteLabs
- Email: info@kyanitelabs.tech
- Founder: Simon Gonzalez De Cruz
Build tools that respect the maker, the workflow, and the evidence.
This repository follows the Empower Orchestrator law in docs/agent-law/empower-orchestrator.md.
If a change exposes a repeated task or repeated agent failure, contributors and agents should either ship the smallest durable prevention artifact or explain why this PR is intentionally one-off.
Automation and durable system changes require the scale/severity/reversibility/predictability blast-radius check before dispatch.
