Ava Supernova
AvaSupernova

Built in the Open

Hoja de ruta

Where Ava has been and where she's going.

88%
Completado
88
Lanzado
12
Próximamente
🧠

Intelligence

27/28
Core agent loop with 90+ tools
7 thinking modes — Code / Plan / Chat / Teach / Security / Brainstorm / Write
24 specialist personas with Conductor orchestration
5-layer memory architecture (extract → reflect → accumulate → analyse → consolidate)
Memory Agent — curated briefs, not raw dumps
Memory hot-path cleanup — durable user/project facts only
Two fleets — Maestro and Supernova, one picker
AutoCoordinator orchestration gate — no task expansion from single-action requests
Self-inspect pipeline — Ava reads her own code from Supabase
10 knowledge packs (game dev, web, mobile, API, DevOps, systems, data science)
Self-improvement vault
Flat system prompt with per-turn mode prefix — 80% token reduction
Listen-first intent gate — Qwen Flash classifier, soft-nudge style
Hard exploration budget — blocks runaway reads before they burn tokens
Compression overhaul — 70% trigger, task continuity preserved post-compress
Tool schema aligned with model conventions (write/read/edit aliases)
Image downsampling + history aging — per-image cost cut 60-80%
Desktop and browser automation in the IDE — open apps, click, type and drive a browser, with a plan you approve first (Windows)
Smart replanning — coordinator replans from blackboard state on failure
Native-agentic dataset capture — 10 datasets, append-only JSONL, ready for the Ava model
Library → Papers — scientific paper explainer with four-layer pass
Native vision routing — images auto-routed to a vision-capable model
Room-scoped assistants — focused Ava per room (Learning / Health / Design), shared memory
Deny a tool call with a reason — Ava adjusts instead of retrying
Stop keeps the conversation, Esc stops Ava
Conversations recover by themselves after a provider refusal
Memories saved when a session goes quiet
MCP support — connect Ava to MCP tools, starting with a live Blender connection
💻

Surfaces

12/13
VS Code extension — published on the Marketplace
CLI agent
Consolidated sidebar — same shape across extension + IDE
Tool-call blocks — top-level stream elements with inline diff views
Context bar — real-time token usage indicator, click-to-compress
Stop button — real hard cancel, momentum cleared
Single-bubble responses
Resizable sidebar with left/right flip
Full-window dashboard layout — every page uses the screen width on wide / 4K monitors
Data portability (export / import)
Ava Supernova IDE — Windows desktop IDE with automatic updates, launching with 1.0
Works without an account — local usage and spend tracking
All 20 languages load properly across the dashboard
🎓

Education

13/15
Teach mode with curriculum generation — 5 specialists (Curriculum Architect, Content Writer, Fact Checker, Quiz Master, Tutor)
Spaced repetition + per-lesson progress tracking
Fact-checked content with web search
20-language support across the agent surface
Game development knowledge pack
AI-era learning paths — learn by building
Community-contributed knowledge packs
Library → Papers — read scientific papers with Ava in plain language
Interactive teach, do and check lessons — graded open input, not just quizzes
Curated Courses hub — Ava-crafted courses with covers, published to your library
Live open grading — Ava marks the answer you actually typed, not a model answer to read against
Curated courses re-authored as interactive steps — teach a bite, you do it, Ava checks
Learning that watches — session logs feed the same shared memory the health loop uses, so lessons adapt instead of resetting
Honest lesson scores, save a course for later, and certificates
Ratings on courses, recipes and exercises
🔒

Privacy & Security

12/12
Local-first architecture — every feature works without an account
All data in ~/.ava/ by default
BYOK users fully private — zero platform reporting
Secret vault for API keys
Security audit mode (OWASP)
Atomic token / credit enforcement
Local-only chat — full agent loop with no cloud round-trip
Local only — your data never leaves your machine
Start over — wipe everything from Settings → Privacy
Account protection — automated and fake signups are stopped before they can use the platform, and free credits go to real people only
Constant watch — suspicious activity is spotted around the clock and contained automatically
Privacy-friendly signups — email alias services like Proton Pass, SimpleLogin and DuckDuckGo are welcome
🚀

Platform & Business

10/10
Web platform with auth + usage tracking
9 BYOK providers, 20+ models — DeepSeek, Kimi, Qwen, Zhipu/GLM, Mistral, MiniMax, Xiaomi MiMo, Tencent Hunyuan and NVIDIA Nemotron
Ava Credits — unified credit-based pricing
Free tier — 300 monthly credits, BYOK unlimited
Paid plans live — Pro $19, Ultra $39, Enterprise $79
Credit top-ups — $3 / $8 / $15 bundles
Credit usage bar with real-time deduction
Period rollover for paid plans
Device sessions — one key for the extension and IDE
Live chat support — Ava first-line triage
🎨

Creative

6/9
Design Studio — guided asset generation
Icon studio — deterministic SVG + AI (Qwen)
Free-form image generation (Qwen)
Video generation (Wan 3.0) — aspect-aware
Voiceover (Qwen3-TTS) — all 48 voices, with translation
Typed asset library + local storage manager
Game Dev asset types — 13 types coming
Web / App asset types — 8 more coming: app icon, badge, avatar, banner, hero, social image, illustration, pattern
3D models with Blender — describe a prop, agree the brief with Ava, and she builds it game-ready in your own Blender, saved to your Library with a 3D viewer
🌱

Health Mode

8/13
860-dish recipe library across 71 regions
170 canonical exercises across 14 movement patterns
Ava-authored technique guides, beginner cues, common-mistakes notes, plus hero and demo imagery per entry
Personalised meal plans that know your pantry, taste history, and training schedule
Adaptive workouts that respond to logged soreness, sleep, energy, and available equipment
Mobility and breathwork sequences for recovery and stress regulation
Log what happened — mark a meal eaten, swapped or skipped and an exercise done or skipped, with a note Ava reads
Weekly logs feeding Ava's memory so recommendations sharpen instead of resetting
Community-shared meal plans and training paths, open dataset
Multi-week meal, fitness and combined plans, saved to your plan library
Shopping list that knows what you skipped
Plans built around the equipment you actually have
Dated plan sessions with movement pictures

Have a feature idea? Open an issue or tell Ava — she can propose tools herself.