Every release, addition, and fix in Idexal Code — tracked in one place so you always know what we've built and what's next.
research skill — ties ingestion together: gather from the web (Crawl4AI) or files (MarkItDown), distil, verify, and save as linked notes in the knowledge graph. Eight bundled skills now.fallback config option (ordered backup models) + tested selection logic, so a task falls back to another model instead of stalling. Request-flow wiring + a Settings UI are next..idexal/memory/*.md notes ([[wikilinks]]). Your project's knowledge builds itself into a browsable graph — open it in Obsidian — syncs through git, and is recalled into every agent. Self-hosted, no subscriptions.loop workflow runs idexal on a schedule to do a recurring task (e.g. a daily triage) with no human prompting. Safe by default — a no-op until you add the IDEXAL_API_KEY secret. Also a copy-paste template for any repo.verify skill (maker/checker) — independent, adversarial verification: confirm a result is actually correct by executing checks, not by reading. The checker the loop leans on.security-review (red-team find + blue-team confirm) and codebase-exploration, distilled from the idexal/skills roles. Seven defaults now./loop command — /loop <goal> runs a goal through the verified-loop discipline. A third trigger alongside Loop mode and the scheduled Actions loop.tdd, code-review, systematic-debugging, loop), no config or network. Project skills in .idexal/skills/ override them by name.loop skill gives agents a verified-loop discipline (discover → record state → implement → verify → gate), built on agent memory and skills. Design the loop once instead of prompting every micro-step.memory tool agents call to save durable facts, stored in the app's local SQLite database..idexal/memory.md (and a global memory.md) are injected too, and you can edit them directly like AGENTS.md.⌘⇧M).api.idexal.com) is the default provider with automatic model discovery; unused providers removed.