Pioneers

AI that teaches itself to use any software.

A generalist skill framework where LLMs autonomously learn to operate applications — browsers, terminals, and anything you can plug in. Give it a task in plain language, and autoskill figures out the interface, executes the steps, and remembers what worked. The next run is faster because the AI already knows how.

• BrowserSkill — AI navigates websites, logs in, extracts data. No selectors, no per-site config.
• TerminalSkill — generates and executes shell commands from natural language with adaptive safety.
• Universal memory — learns from every run. First execution explores, second execution flies.
• Destructive command blocker — keyword + LLM two-layer safety gate for terminal operations.
• Adaptive trust — new tools start in sandbox mode, graduate to live execution after 3 successes.
• Plugin system — add custom skill backends in under 50 lines of Python.
• Unified CLI — auto-detects whether a task needs a browser or terminal.
• Streaming execution — watch the AI work in real-time, intervene when needed.
• Checkpoint recovery — save and restore execution state on failure.
• Self-evolution — skills learn from failures and improve autonomously.

Never miss an upstream change again.

A CLI that brings structure to tracking upstream repositories. Observe commits, PRs, and releases across GitHub, GitLab, Gitea, gitcode.com, and Codeberg. Classify changes automatically, assess relevance against your project, generate adaptation plans with risk scoring, and contribute improvements back — all from the terminal.

• Multi-provider — GitHub, GitLab, Gitea, gitcode.com, Codeberg, and self-hosted instances.
• Structured lifecycle — observed, analyzed, assessed, planned, implemented, validated, contributed.
• Auto-classification — security fixes, bugfixes, refactors, and features sorted by regex + heuristics.
• Relevance scoring — risk assessment against your project's modules and policies.
• Adaptation plans — concrete strategies with implementation branches and draft PRs.
• Git-tracked state — all data lives in .adapt/ as YAML and JSON. No database, no server.
• Dual output — rich terminal tables for humans, --json for scripts and CI.
• Xinchuang compliance scoring — check stack compatibility against approved technology lists.

Your codebase, searchable in seconds.

A developer knowledge base that scans hundreds of git repositories, analyzes languages and patterns, generates AI summaries with local Ollama, and lets you find anything with hybrid search. Fulltext for exact matches, semantic for meaning, pattern-based for architecture — fused with Reciprocal Rank Fusion into one ranked result.

• One command scan — discovers repos, detects 70+ languages, maps dependencies and patterns.
• AI summaries — local Ollama generates descriptions and tags for every project.
• Three-engine hybrid search — SQLite FTS5 + LanceDB vectors + pattern matching, fused with RRF.
• Confidence scoring — every result tells you how reliable it is.
• Smart chunking — title-aware paragraphs with TF-IDF scoring and sliding context windows.
• Heuristic reranker — keyword density, freshness boost, and code quality signals.
• Plugin system — extend scanning and analysis with custom plugins.
• Works offline — AI features use local models. Degrades gracefully without Ollama.
• Incremental indexing — only re-processes projects that changed since last scan.

Turn feature descriptions into structured specs, plans, and tasks — visually.

A visual interface for spec-driven development. Write a feature description in plain text, and spec-intelligence generates a structured specification, clarifies ambiguities interactively, produces an implementation plan, and breaks it into tasks on a kanban board. Works as a web app or desktop app — your specs stay as Markdown files, fully compatible with CLI workflows.

• Spec authoring — create structured specifications from plain-text feature descriptions with a rich hybrid editor.
• Pipeline visualization — see the full lifecycle (specify → clarify → plan → tasks → implement) for every feature at a glance.
• Kanban task board — drag-and-drop task management with phase grouping, dependency arrows, and parallel-execution markers.
• Interactive clarification — resolve [NEEDS CLARIFICATION] markers through guided dialogs with suggested options.
• Plan generation — produce implementation plans with architecture decisions, data models, and interface contracts via local LLM.
• Extension system — browse, install, and manage extensions and presets from configurable catalogs.
• Desktop + web — runs as a Tauri desktop app for local use or a self-hosted web app for team collaboration.
• File-first — all specs live as Markdown in your repo. Full CLI compatibility, git-tracked, no database.
• Cross-artifact analysis — detect orphaned requirements, missing coverage, and constitution violations across all project specs.
• GitHub issue export — preview and push tasks as GitHub issues with labels, dependencies, and bidirectional linking.