SkeletIQ is the architecture layer for AI coding agents. Describe a system in a prompt, and it comes back as a critiqued architecture on an editable canvas: components, data stores, connections, drawn out and immediately scored.

The scoring is not a second opinion from a model. A deterministic pass over the component graph runs a rule-based critique (fan-out, sync-chain depth, blast radius, missing trust boundaries), a traffic and capacity simulation that finds the first component to saturate, cloud cost estimates across three providers, a design-time compliance lint against SOC 2, HIPAA, PCI-DSS, GDPR and SOX, and a set of architecture decision records. You edit on the canvas, not just accept what came back, then release a version. Releases are immutable.

From there, an MIT-licensed MCP server hands the release to your coding agent: a brief written into your repository's AGENTS.md, a build order to work through component by component, and open questions the agent is not permitted to answer on its own. After the build, check_drift reports four buckets: built, half-built, missing, and not-in-design, the last one being what nobody decided on.

Generation is LLM-driven. The analysis on top, critique, drift, build order, decision records, is derived from the graph and deterministic. Compliance findings and inferred scale are advisory and founder-reviewed, not audit evidence.

Two editions from one codebase: a hosted SaaS and a self-hostable edition planned. Built and run by one person, Sabhahith Works Private Limited, in public beta.