Most problems are one of six things, and the troubleshooting page lists them in the order they happen. What follows is everything else.
Install to first reply.
Six fixes, in order.
What a seat is, and what each plan unlocks.
Keys stay in your editor’s secret storage (or, on the web, in your browser or your Key Vault as ciphertext we cannot read). The extension makes no inference proxy calls. When you use Server Proxy (paid plans, or a free account for a provider that blocks browser calls), your key passes through our server for that request and is not stored. The extension displays reported usage and estimated cost locally; your provider bills the API usage.
Only to the providers you seated, as part of the prompts you trigger. /context reports token and price estimates for the prepared provider payload, and /context full includes the exact system prompt; /context save compacts repeated history into a Project-owned CONTEXT.md.
Start with a Drafter for a first reply, then add a Reviewer for collaboration. A third seat can add context or a tiebreaker when your configured capacity permits. A formal Project is optional.
Yes. You can assign models from one provider to different roles. You can also mix providers to bring different perspectives to the same task.
1.95.0 and newer (`engines.vscode` ^1.95.0).
The tested installation path is VS Code desktop. Compatibility with other editors and remote environments needs separate verification; do not assume every VS Code-compatible editor exposes the same APIs.
They are files in your repo. PLAN.md, EXECUTION.md and memory/ stay exactly where they are, and the extension keeps working at Basic limits.
Include your VS Code version, AI Orchestra build fingerprint, the action you tried and the error message. Remove keys, account tokens and private project content before sharing diagnostics.
Email support@ai-orchestra.app. For privacy questions, contractual matters, or general inquiries, contact info@ai-orchestra.app.