{"slug": "ava-supernova-open-source-coding-agent-for-open-weight-llms", "title": "Ava Supernova – open-source coding agent for open-weight LLMs", "summary": "Ava Supernova, an open-source AI coding agent, launches with support for open-weight LLMs, offering 60+ tools, 7 modes, and 24 specialist personas. The agent is free to start, local-first, and requires no account or credit card, aiming to democratize agentic coding by running on frontier open-weight models that match closed models at lower cost.", "body_md": "**The open-source AI coding agent that works for everyone — not just the people who can afford it.**\n\n60+ tools · 7 modes · 24 specialist personas · frontier open-weight models from every major lab · a Creative Studio that builds real assets into your project. **Free to start, no account, no credit card.**\n\n[What Ava Is](#what-ava-is) ·\n[Surfaces](#one-ava-four-surfaces) ·\n[Models](#models--fleets) ·\n[Getting Started](#getting-started) ·\n[Tools](#tools) ·\n[Privacy](#privacy) ·\n[Contributing](#contributing)\n\nAgentic coding shouldn't be a luxury. Ava Supernova is a full AI coding agent — she reads your codebase, plans the work, executes it task by task, and remembers what she learns so the next time gets easier. She runs on the frontier of **open-weight** models, which now match or beat closed frontier models on agentic coding at a fraction of the cost.\n\nShe's open-source, **local-first**, and free to start:\n\n**No account, no credit card, no trial.** Install, open, go. Bring your own API key and run fully local — or sign up for**300 free credits a month** and start with zero setup.**Your data stays on your machine.** Memory, history, journal, tasks — all local by default. Cloud sync is opt-in.**No gatekeeping.** Every capability — including the security toolkit and the teaching engine — is available to everyone. The free tier is a floor, not a free trial.\n\nAsk Ava\n\n`what can you do?`\n\nand she'll read her own source code and tell you herself.\n\nThe same agent engine (`@ava/core`\n\n) powers every surface — same tools, same personas, same memory. You choose where you work:\n\n| Surface | What it is | Best for |\n|---|---|---|\nCLI (`@ava/cli` ) |\nA terminal REPL with the full agent loop. | Living in the terminal, scripting, servers, SSH. |\nVS Code Extension |\nAva inside the editor you already use — chat, agent, command-centre dashboard. | Adding Ava to your existing VS Code workflow. |\nIDE |\nA standalone desktop app (Tauri + Rust) with Ava native and a local engine. Owns desktop automation. |\nA dedicated AI-native workspace; controlling the whole machine. |\nCompanion |\nA mobile/web app — chat, tasks, journal, memory on the go. | Ava away from your desk. |\n\n**She remembers you.** A layered memory system learns your style, conventions, decisions, and the bugs you've fixed before — curated by a dedicated Memory Agent so context stays lean and recall stays sharp. Local-first, opt-in semantic recall.**She thinks before she builds.** 24 specialist personas across 7 modes plan, verify, and challenge before executing. Simple questions skip orchestration entirely — zero overhead.**She picks the right fleet for the work.** You pick the job; the right multi-model fleet falls out of it (see[Models & Fleets](#models--fleets)).**She teaches for free.** Switch to Teach mode, say \"teach me Rust,\" and she builds a personalised curriculum and teaches it lesson by lesson — fact-checked against the live web. Same credits as every other mode.**She explains herself.** Ask \"how does your memory work?\" and she reads her own implementation and walks you through it.**She's yours to design.** Name, pronouns, tone, energy, communication style — same brain, your personality. Full UI in 20 languages.\n\nModes aren't tool restrictions — they're states of mind. The toolkit and the persona team change to match the work.\n\n| Mode | Prefix | Mindset | Use it for |\n|---|---|---|---|\nCode |\n`>>` |\nBuilder | Writing, editing, shipping code with the full toolkit |\nPlan |\n`::` |\nArchitect | Read-only analysis, feature design, strategic proposals |\nChat |\n`..` |\nFriend | Personal conversation, journal, memory, news, weather |\nTeach |\n`??` |\nTutor | Personalised learning paths and Socratic delivery |\nSecurity |\n`!!` |\nAuditor | OWASP-aligned scanning with verified findings |\nBrainstorm |\n`**` |\nIdeator | Research, generate, challenge, and refine ideas |\nWrite |\n`<<` |\nAuthor | Drafting real documents — reports, proposals, letters, articles — exported as branded Word & PDF |\n\nSwitch with the dropdown or `Ctrl+Shift+1`\n\nthrough `Ctrl+Shift+7`\n\n.\n\nWhen a task is complex, Ava's internal team activates. Each persona gets its own clean context and scoped tools (the Challenger reads but never writes). Simple tasks go direct.\n\n**Code:** Scout → Architect → Verifier → Sequencer → Challenger → Builder**Plan:** Researcher → Architect → Challenger**Teach:** Curriculum Architect → Content Writer → Fact Checker → Quiz Master → Tutor**Security:** Recon → Scanner → CVE Researcher → Verifier → Reporter**Brainstorm:** Explorer → Researcher → Ideator → Challenger → Refiner\n\nChat and Write run without personas — just Ava.\n\nYou don't pick a model — you pick the work, and the right fleet falls out of it. Two orchestrated fleets ship today; a third is rolling out.\n\n| Fleet | Status | Coordinator | Best for |\n|---|---|---|---|\nMaestro |\nShips now |\nQwen 3.7 Plus | Daily work, predictable cost — a tier-differentiated Qwen ensemble |\nAurora |\nShips now |\nMistral Medium 3.5 | GDPR-strict / sovereign EU work — Mistral-only, open weights, never leaves EU infrastructure |\nSupernova |\nBYOK now · managed preview | DeepSeek V4 Pro | Heavy multi-step work — a polyglot fleet picking the best model per role |\n\n**Supernova** runs today with your own DeepSeek + Qwen keys; the managed (platform-hosted) version is in preview. **Maestro** unlocks with a Qwen key, **Aurora** with a Mistral key.\n\nBring your own key for any provider, or use Ava-managed models on a plan. Every plan gets every managed model — tiers differ only on monthly credit allowance, never on which models you can use.\n\n| Provider | Models |\n|---|---|\nAva Managed |\nQwen 3.7 Plus, Qwen 3.5 Plus, Qwen 3.5 Flash, plus the managed Mistral (Aurora) coordinators |\nQwen (Alibaba) |\nQwen 3.7 Plus, Qwen 3.7 Max, Qwen 3.5 Flash — 1M context, vision, reasoning |\nDeepSeek |\nV4 Pro, V4 Flash — MIT open-weight, 1M context |\nMoonshot |\nKimi K2.7 Code — 1T MoE / 32B active, 256K context |\nZhipu |\nGLM-5.2 — MIT open-weight, 1M context |\nMistral |\nMedium 3.5, Small 4, Large 3, Codestral, Devstral — EU, open weights, up to 262K context |\nMiniMax |\nM3, M2.7 — 1M context (BYOK) |\nXiaomi |\nMiMo V2.5, V2.5-Pro — 1M context, native multimodal |\nAnthropic |\nClaude Opus 4.8, Sonnet 4.6, Haiku 4.5 |\nCustom / Local |\nOllama, LM Studio, vLLM, or any standard API-format endpoint — the $0 path |\n\n[Node.js](https://nodejs.org)v20.0.0 or later- An API key from any supported provider —\n**or sign up for 300 free credits a month**\n\n```\n# Install from npm\nnpm install -g @ava/cli\nava\n```\n\nOn first launch, Ava runs an interactive setup wizard: pick a provider, paste your key (or skip and sign in for the free tier), and you're ready.\n\nInstall from the [VS Code Marketplace](https://marketplace.visualstudio.com/items?itemName=augmentedvalueacceleration.ava-supernova) — or search \"Ava Supernova\" in the Extensions panel. Click the Ava icon in the activity bar (or `Ctrl+Shift+A`\n\n) and start.\n\nDownload the standalone desktop IDE from [ava-supernova.com](https://ava-supernova.com). A local Node.js engine runs `@ava/core`\n\non your machine — the full agent without a browser or VS Code.\n\n```\ngit clone https://github.com/AugmentedValueAcceleration/ava-supernova.git\ncd ava-supernova\npnpm install\npnpm build\n```\n\n60+ user-facing built-in tools. Ava decides which to use, runs them, reads the results, and keeps reasoning. (The IDE adds desktop- and browser-automation tools on top, gated to Desktop mode.)\n\n| Category | Tools |\n|---|---|\nFile ops |\n`file_read` , `file_write` , `file_edit` , `glob` , `grep` , `list_directory` , `find_symbol` , `project_index` , `analyze_architecture` , `self_inspect` , `docs_lookup` , `browse_library` |\nShell |\n`bash` , `test_run` , `test_generate` , `benchmark` , `debug_logs` |\nGit |\n`git_status` , `git_diff` , `git_commit` , `git_create_pr` , `rollback` |\nWeb |\n`web_search` , `http_request` , `browser` , `weather` , `news` , `release_notes` , `paper_fetch_full_text` |\nMedia |\n`generate_image` , `generate_video` , `remove_background` |\nDocuments |\n`document_author` , `document_manage` , `email_draft` , `report_generate` , `doc_generate` , `present_plan` , `apply_plan` , `todo_write` , `task_manage` , `journal_write` , `switch_mode` |\nMemory |\n`memory_save` , `memory_recall` , `memory_update` , `memory_delete` , `curator` |\nLearning |\n`learning_create` , `learning_teach` , `learning_progress` |\nSecurity & Data |\n`audit_dependencies` , `database_query` , `secret_request` , `env_write` |\nSystem |\n`get_datetime` , `detect_language` , `ask_user` , `support_request` , `propose_tool` |\n\nAva generates real assets and wires them straight into your project — a logo into `public/`\n\n, a hero clip into your landing page.\n\n**Images & video** run on**Wan**(Alibaba's open-weight models), built for clean graphic-design output and short clips with synchronised audio.- Every asset is tracked in the\n**Library**— grid or list view, filter by type, preview, reuse across projects.\n\nYour data stays yours:\n\n**API keys** encrypted in your OS keychain (VS Code SecretStorage). Never logged, never sent to Ava servers.**Conversations & memory** stored locally (`~/.ava/`\n\n). Local-first by default.**Credential safety**— runtime detection blocks API keys, tokens, and private keys from being saved to memory; high-confidence key patterns are redacted from chat output before they hit the screen.**Capability-based secrets**— Ava receives opaque handles for vault secrets, not raw values, scoped to the current session.** Prompt-injection resistance**built in.** No telemetry**— we don't track your usage, code, or conversations.** Cloud sync is opt-in**, with delta tracking so you see exactly what's about to be pushed.\n\nEvery plan gets every managed model. Tiers differ only on monthly credit allowance.\n\n| Tier | Price | Monthly credits |\n|---|---|---|\nFree |\n$0 | 300 |\nPro |\n$19 | 5,000 |\nUltra |\n$39 | 10,000 |\nEnterprise |\n$79 | 20,000 |\n\nOr run fully local with your own keys — the floor is free, forever.\n\nA pnpm monorepo with a shared core, plus submodules for the IDE, web platform, and companion app:\n\n```\npackages/\n├── core/        # @ava/core — shared agent engine (agent loop, providers, tools, personas, memory)\n├── cli/         # @ava/cli — terminal REPL\n├── extension/   # VS Code extension host + React webview + dashboard\n├── ide/         # Standalone IDE (Tauri v2 + Rust + Node.js engine) — submodule\n├── mobile/      # Companion app — submodule\n└── web/         # Platform website — submodule\n```\n\n**Key design decisions:**\n\n**Shared core, zero duplication**— CLI, extension, and IDE all run the same`@ava/core`\n\nengine, tools, and providers.**Event-driven agent**— the agent emits typed events every host subscribes to.** Self-contained tools**— each tool carries its own schema and handler; adding one is a single file plus one registry line.** No provider SDKs**— all provider communication uses native`fetch`\n\n.**Permission matrix**— risk levels (`safe`\n\n,`write`\n\n,`dangerous`\n\n) combine with your permission mode to decide what needs approval. Plans and user questions always require approval.**Local-first**— everything works offline; cloud sync is opt-in.\n\nContributions are welcome — this is built for the community.\n\n- Fork the repository\n- Create a feature branch from\n`development`\n\n- Make your changes and run\n`pnpm build`\n\n- Open a pull request against\n`development`\n\n[Apache License 2.0](/AugmentedValueAcceleration/ava-supernova/blob/development/LICENSE) — Copyright 2026 Augmented Value Acceleration\n\nAgentic coding for everyone.", "url": "https://wpnews.pro/news/ava-supernova-open-source-coding-agent-for-open-weight-llms", "canonical_source": "https://github.com/AugmentedValueAcceleration/ava-supernova", "published_at": "2026-06-26 15:38:06+00:00", "updated_at": "2026-06-26 16:05:25.030754+00:00", "lang": "en", "topics": ["ai-tools", "ai-agents", "developer-tools", "large-language-models"], "entities": ["Ava Supernova", "VS Code", "Tauri", "Rust", "OWASP"], "alternates": {"html": "https://wpnews.pro/news/ava-supernova-open-source-coding-agent-for-open-weight-llms", "markdown": "https://wpnews.pro/news/ava-supernova-open-source-coding-agent-for-open-weight-llms.md", "text": "https://wpnews.pro/news/ava-supernova-open-source-coding-agent-for-open-weight-llms.txt", "jsonld": "https://wpnews.pro/news/ava-supernova-open-source-coding-agent-for-open-weight-llms.jsonld"}}