# Getting started with Mysti 2.0 BETA ## Install and open Use VS Code 1.86 or newer and a trusted local workspace on Windows, macOS, or Linux. Install `DeepMyst.mysti` from Extensions. For an unpublished beta, use **Extensions: Install from VSIX** with the reviewed package. Reload the window after replacing an installed version. Open **Mysti: Open Chat** from the command palette or click Mysti in the Activity Bar. Opening a separate editor tab gives a larger workspace for comparisons and visual work. ## Configure one agent Choose an agent in the composer and follow the setup screen. Installation, CLI discovery, authentication, and model support are separate checks: completing an installer does not imply sign-in is complete. - **Claude Code / Codex / another CLI:** use the provider's install and authentication actions. If Mysti detects a model compatibility error requiring a newer CLI, use the upgrade action and wait for discovery to refresh. - **Mysti:** sign in to DeepMyst. The coordinator uses that account; delegating to a local CLI still requires that CLI's setup. - **Ollama / LocalAI / OpenRouter:** configure the appropriate endpoint or account. These are not all installed through npm. See [Providers](PROVIDERS.md) for adapter-specific details. Model catalogs and account entitlements can differ from examples in screenshots. ## Assign a task ```text @claude Explain this function @claude @codex What are your opinions on this design? @claude Write the parser, then @codex review it @claude:critic @codex:reviewer Assess this proposal ``` The selected provider does not absorb explicit assignments. Opinion requests have separate, concurrent read-only runs. A `then` handoff waits for the preceding assignment. Work that may write files runs serially and retains your permission settings. Add files through Context or `@path/to/file`. Explicit collaborator runs currently do not forward composer attachments; the UI warns you to use Context. Dictation inserts text into your draft and does not send it automatically. ## Read the results Each live assignment card identifies its agent and status. The final conversation stores attributed answers so reopening it preserves who said what. A missing, unauthenticated, timed-out, or failed participant is reported rather than replaced. Stop cancels active collaborators; a new send supersedes the previous run. `@agent:role` controls perspective and access. Unknown or unverified roles do not gain write authority. See [Mentions](MENTIONS.md). ## Try connected context Open **Mysti: Open Proactive Inbox**. Local Git monitoring needs a chosen workspace and the editor running. Cloud GitHub/Slack monitoring needs DeepMyst sign-in, a supported connected account, and a responsibility you explicitly create. Review evidence before acting on it; author metadata is not task ownership. [Setup and limits](PROACTIVE.md). ## If something goes wrong | Symptom | Check | | --- | --- | | An agent is unavailable | Its setup card: installed path, CLI version, and sign-in are distinct checks. | | A model is rejected | Choose a model your account supports; use an offered CLI upgrade if required. | | Only one opinion arrives | Inspect the other agent's card and error. Retry after fixing its setup. | | Work runs serially | Writers and explicit dependencies serialize. Independent read-only work respects `mysti.collab.maxConcurrent`. | | Dictation has no microphone support | Follow the editor's speech setup prompt and OS microphone permissions. | | DeepMyst sign-in fails | Retry sign-in, inspect the reported error, and include the HTTP status in a bug report. Never share tokens. | Report your Mysti version, OS, VS Code version, provider/CLI version, exact prompt with private content removed, and expected versus actual behavior. Attach a redacted screenshot or recording where helpful.