The full agent, in your editor.
Seven modes of thought. A conductor of specialist personas. Five-layer memory that carries your decisions across sessions. Run it on any frontier model — or free, fully offline, on your own local stack. No account required to start.
Seven modes of thought
Switch how Ava thinks with a two-character prefix. Each mode is a different mindset — different tools, different priorities, different voice.
Code
Builder mindset. Read, write, run, commit, ship — the full toolset behind one intent.
Plan
Architect mindset. Read-only. Designs the approach before a line is written.
Chat
Friend mindset. Conversation, memory, journal. No code unless you ask.
Teach
Tutor mindset. Step-by-step, grades your real answers. Guides over gives.
Security
Auditor mindset. OWASP review, dependency scanning, threat modelling.
Brainstorm
Ideator mindset. Research-grounded, challenges weak ideas, personal to you.
Write
Author mindset. Compose real documents in Markdown — export to branded Word and PDF.
The Conductor
Orchestrates the specialist personas behind each mode — planning first, always.
Bring any brain — or none at all
Plug in your own key for any frontier model, route through a managed tier, or point Ava at a model on your own machine and pay nothing, ever.
Managed routing
Maestro, Supernova, Aurora — pick a tier, Auto picks the model. 300 free credits a month, no card.
Your own key (BYOK)
Any frontier model on your own account. Keys live in your OS keychain — never our servers, never logged.
Local & free
Point at a model on your machine — the full agent, $0, fully offline. The floor is free, not a free trial.
She plans before she touches a line
In Code mode, specialists run in a wave — gather, design, challenge, sequence, build. Independent ones run in parallel; the risky work waits its turn.
Scout
Gathers context
Reads files, indexes the project, understands what exists before suggesting what should.
Architect
Designs the plan
Structures the approach — file by file, step by step. No surprises.
Verifier
Validates the plan
Hunts conflicts, edge cases, coverage gaps. Catches mistakes before they happen.
Sequencer
Orders the work
Dependencies first, risky changes last. The safest path through.
Challenger
Stress-tests
Asks "what if?" — pushes back on weak plans, forces better thinking.
Builder
Executes
Only now does code get written. Confident, tested, verified.
A real toolset, built in
Not plugins bolted on — native capabilities that work together inside the editor.
File operations
read, write, edit
Search
glob, grep, symbols, index
Shell
bash, tests, benchmarks
Git
status, diff, commit, PR, rollback
Web
search, HTTP, opt-in browser
Creative Studio
image, video, music, voice
Memory
save, recall, update, delete
Authoring
documents, Word & PDF export
Planning & tasks
plans, todos, task management
Journal
you + Ava, mood tracking
Learning
curriculum, teach, progress
Health
recipes, movements, plans
Security
OWASP audit, dependency scan
Architecture
codebase analysis, self-inspect
Office
emails, reports
Awareness
weather, news, release notes
One thing the editor can’t do
VS Code runs the extension in a sandbox — Microsoft’s rules, not ours. So Ava can’t take screenshots or operate other apps from here, and we won’t pretend otherwise. When you need her to drive the whole machine, that lives in the desktop IDE — same Ava, no sandbox.
She can read her own source code
Ava’s self_inspect tool reads her own implementation, straight from the repository — no rate limits, no API calls. She can explain how she works and help contributors find the right file. Transparency isn’t a feature here; it’s the foundation.