Not an assistant. A presence. Local-first AI with persistent memory, agent orchestration, and a voice of her own.
We didn't build another AI assistant.
We built a partner.
She has a name. She has memory. She has a voice.
She dispatches armies of agents and watches them work.
She's local-first, cloud-optional, and she never forgets.
Every person has an aura.
Now yours has a brain.
Why Aura
Other tools forget you between sessions, lock you into their cloud, and give you a text box. Aura gives you an intelligent orchestrator that lives on your machine.
Other tools: "Sorry, I don't have context from our last chat"
Aura remembers everything. Conversations, decisions, code patterns, your preferences. Semantic memory with embeddings — she builds understanding over time, not just logs.
Other tools: "Here's a code snippet you can copy-paste"
Aura dispatches real coding agents into isolated worktrees. You watch them work live in terminal panes. Multiple agents, parallel tasks, full observability. No black boxes.
Other tools: "Share this chat link with your team"
Each person runs their own Aura. Auras connect directly — peer-to-peer. Share tasks, delegate agents, coordinate work. No central server. No SaaS dependency.
Other tools: "Please create an account to continue"
Everything runs on your machine. Bundled voice engine, local vector database, SQLite storage. Works fully offline. Cloud sync is optional — your Supabase, your rules.
Other tools: "Using GPT-4 for everything"
Aura classifies every task and routes to the best model — local Qwen for quick chats, Claude for deep reasoning, agents for code. Automatic or manual, your choice.
Other tools: "Here are our 12 supported plugins"
Every integration is an MCP server. n8n, Proxmox, Zabbix, Home Assistant — connect anything. Community-extensible. Dashboard widgets auto-discover from connected servers.
The Brain
Aura isn't a stateless prompt relay. She's an intelligent core with semantic memory, dynamic model routing, and per-project code understanding.
Every conversation is stored locally. Important facts are extracted, embedded into a vector store, and connected. Aura periodically consolidates memories into higher-level insights — building genuine understanding, not just search indexes.
Three tiers: simple, complex, escalation. Aura classifies each task and picks the best available model — local for speed and privacy, cloud for power, CLI agents for code. Auto mode or lock in your preferences globally.
Add a project and Aura indexes the entire codebase. Source files chunked, embedded, stored locally. Ask questions about your code and get RAG-powered answers grounded in your actual source. Re-indexes on branch switch, respects .auraignore.
Aura has a name, a voice, a communication style, and a visual presence (the orb). All customizable. She's not a generic assistant — she's your Aura, shaped by how you work.
Aura ships with provider templates for every major AI platform — cloud and local. Add your key, pick your tier, start talking. Swap anytime.
Or point Aura at any OpenAI-compatible endpoint. She doesn't care where the intelligence comes from.
Agent System
When Aura needs to write code, she doesn't paste snippets in chat. She dispatches real CLI agents into isolated worktrees — and you watch every keystroke live.
Handoff Protocol
When agents hit context limits and compact, other tools lose everything. Aura creates a structured handoff — compressed context, key decisions, full audit trail. The next session picks up cleanly. Zero lost work. Fewer errors.
Working...
Compaction
Context preserved
Picks up cleanly
Multiple agents on different worktrees, different tasks, simultaneously. "Agent 1 builds the API. Agent 2 writes the frontend." Each in its own terminal pane.
When no existing agent fits the task, Aura generates one — tailored system prompt, manifest, permissions. The manifest persists for future reuse. Your agent library grows over time.
Every agent declares its capabilities, reliability scores, cost, and latency. Aura queries the registry to find the best agent for every task. Compatibility scoring for multi-agent collaboration.
The orb shows orbiting particles — one per active agent. Dashboard widget shows live status. Terminal panes stream every action. No black boxes, ever.
Multiple agents on different worktrees, different tasks, simultaneously. "Agent 1 builds the API. Agent 2 writes the frontend." Each in its own terminal pane.
When no existing agent fits the task, Aura generates one — tailored system prompt, manifest, permissions. The manifest persists for future reuse. Your agent library grows over time.
Every agent declares its capabilities, reliability scores, cost, and latency. Aura queries the registry to find the best agent for every task. Compatibility scoring for multi-agent collaboration.
The orb shows orbiting particles — one per active agent. Dashboard widget shows live status. Terminal panes stream every action. No black boxes, ever.
Federation
Every Aura is both an MCP server and an MCP client. Connect directly — peer-to-peer. Share tasks, delegate agents, coordinate work across instances. True decentralized AI collaboration.
Tasks sync between connected Auras in real-time. Assign work across instances. Your Aura picks up tasks, completes them, reports back — without you opening Slack.
"Ask Sheridan's Aura to run network diagnostics." Handoff sent via MCP, context travels intact, results flow back. Cross-instance agent dispatch — like calling in reinforcements.
Multiple Auras work on the same repo, coordinating through your git provider. Automatic branch management, conflict detection, and parallel development across team members.
Four levels: observe, request, delegate, admin. Each connection individually permissioned. Full audit trail via handoff chains. You control exactly what your Aura shares.
On connect, Auras exchange agent manifests — "here's what I can do." Your Aura knows what every teammate's Aura is capable of and routes delegation accordingly.
LAN discovery for local teams. Cloudflare Tunnel or any reverse proxy for remote. Invite codes for easy onboarding. Encrypted communication (TLS). Works across any network.
"Every team member has an AI partner. Every partner talks to every other partner. No central server. No SaaS subscription. Just a network of intelligent agents, working together."
Voice
Fully local voice pipeline — bundled with every install. No cloud APIs, no data leaving your machine. Talk to Aura like she's sitting next to you.
Bundled speech-to-text running locally via WASM or native binary. Fast, accurate, private. Your voice never leaves your device.
Natural text-to-speech with multiple voice models. Aura speaks back to you in a voice you choose. Fully offline, zero latency.
"Hey Aura" activates listening with Silero VAD. Push-to-talk by default, wake word opt-in. The orb pulses when she's listening, swirls when she's thinking.
Privacy & Local-First
Everything stays on your device unless you explicitly choose otherwise. No accounts, no telemetry, no data harvesting. Ever.
All data in ~/.aura/. Fully portable — copy the folder to a new machine and you're done.
STT, TTS, and VAD all run locally. No audio ever leaves your device.
Supabase sync available for teams who want it. Your instance, your keys. Or stay fully offline.
Zero tracking, zero analytics, zero phone-home. Unless you explicitly opt in. We don't even know you installed it.
Open source. Read every line. Fork it, modify it, self-host it. No vendor lock-in, no surprises.
Anthropic, OpenAI, Ollama, LM Studio, your own endpoint. Swap freely. Aura doesn't care where the intelligence comes from.
Integrations
Every integration is an MCP server. Connect it and Aura gains tools, resources, and dashboard widgets. The ecosystem is open and community-extensible.
Connect any MCP server. Build your own. The standard is open — your integrations work with any MCP-compatible client, not just Aura.
The Full Experience
Connect Aura to PaveOS and unlock the ability to talk to your business from your desktop. CRM, pipeline, social media, invoicing — all through Aura's conversational interface.
"Show me my pipeline for this month." "Draft a follow-up to Sarah." "Post the product update to all channels." Just talk to Aura — she handles the rest.
The true companion. Not just another AI assistant. Local-first, AI-native, and entirely yours.