Scylla

Support

Get Scylla ready for your work.

Set up your Windows build, connect a provider, choose the projects and reference material available to agents, and configure the rules for execution.

Start here and keep working.

Install Scylla on WindowsChoose a Windows build, review its requirements, and complete the first launch.Set up your first workspaceAdd project and reference folders, connect a provider, and inspect a first task.Configure Security PolicySet Allow, Ask, and Block decisions for Scylla-managed operations.Accounts and sign-inUnderstand Windows identity, Scylla licensing, and provider authentication.Invite people to your teamInvite the correct account and manage pending organization invitations.Assign Team seatsAllocate purchased seats and activate Team entitlement on the desktop.Set a managed team policyPublish organization rules and verify the effective desktop policy.Set up Team MemoryConfigure shared project context and distinguish empty results from unavailable service.Manage Team billingReview plan, purchased seats, subscription state, and payment actions.Activate Team on the desktopConfirm membership, seat assignment, and organization configuration.Troubleshoot ScyllaIdentify the failing layer and retry the smallest task that reproduces it.Report a Scylla issuePrepare a useful report without including credentials or sensitive values.Terminal and SSHRun local and remote work through configured, policy-controlled execution paths.Jobs and payload trackingInspect execution progress, output, and tracked provider-bound context.Keyring and connectionsStore reusable credentials locally and use protected references in configured connections.SafeGuardReview potential secret exposure before repository changes are shared.Ask, Plan, and ExecuteUse Scylla's workflow modes without changing the authority boundary.Organize conversationsFile project discussions so earlier reasoning remains useful.Manage supported toolsInspect installed integrations and use maintenance actions available in your build.Plugin developmentExtend Scylla through documented SDK interfaces and execution boundaries.

FAQ

01What is Scylla?

Scylla is a local-first AI development workbench that gives supported coding Agents a controlled place to work with your projects, terminal, tools, connections, and project context. Scylla is not an AI model; you bring the provider accounts and models you already use, while Scylla manages the working environment around them.

Learn more: First Launch →

02Does Scylla replace my ChatGPT, Claude, Cursor, or other AI subscription?

No. Scylla does not resell AI inference or replace your provider account. Provider authentication stays separate from your Scylla Account and Scylla licensing, so you connect supported providers using the accounts and access you already have.

Learn more: Account & Sign-In →

03Can Scylla work with more than one project at a time?

Yes. Every project added to Workspace Scope is active in the current Scylla session, and Agents may work across those active projects. One project can be marked Primary, but Primary is only the fallback when a chat or operation does not otherwise identify its project.

Learn more: First Launch →

04What is the difference between Workspace Scope and Knowledge Scope?

Workspace Scope contains project/code folders Scylla makes available for normal development work. Knowledge Scope is a separate place for plans, architecture notes, handoffs, research, and other non-project reference material; Knowledge can be used and maintained by Agents without being treated as project source or included in normal project Git operations.

Learn more: First Launch →

05Does adding a folder to Scylla change my Windows permissions?

No. Adding a folder to Workspace Scope or Knowledge Scope makes it active inside the Scylla session; it does not replace Windows filesystem permissions. The signed-in Windows user still determines what the application can actually read or write on the machine.

Learn more: First Launch →

06What do ALLOW, ASK, and BLOCK mean?

They are Scylla's Policy verdicts for supported operations. ALLOW lets an operation proceed without a Policy prompt when the other required boundaries pass, ASK stops at that boundary for approval, and BLOCK refuses the operation.

Learn more: Configure Policy →

07Does Scylla sandbox AI Agents?

Scylla provides application-level project scope, Policy, brokerage, and credential controls, but it is not an operating-system sandbox for arbitrary untrusted code. Windows permissions, Git permissions, database permissions, cloud IAM, and remote-service permissions remain real security boundaries and should still follow least-privilege practices.

Learn more: Configure Policy →

08Can an Agent read the passwords and API keys in my Scylla Keyring?

Scylla's credential model is designed around references and brokered operations rather than giving the Agent a reusable secret value. The operation gets the credential; the Agent does not. Unlocking the Keyring makes protected values available to trusted Scylla execution paths, not to Agent chat as a vault export.

Learn more: Keyring & Connections →

09Why can something be connected but still unavailable to an Agent?

Connection and authority are separate in Scylla. A provider can be signed in, an MCP server can be connected, or a database Connection can test successfully while project scope, Policy, credentials, or another capability boundary still prevents a particular operation.

Learn more: Configure Policy → · Troubleshooting →

10Does Scylla upload my whole project to an AI provider?

Scylla is a local desktop workbench and your project remains on your machine. When you use a remote AI provider, the provider still receives the prompts and project context required for the request; adding a project to Scylla does not mean the entire repository is automatically uploaded as one payload.

Learn more: Jobs & Payload Tracking →

11What is Scylla Memory, and how is it different from Knowledge?

Knowledge is material you deliberately place in Knowledge Scope, such as plans, research, blueprints, and handoffs. Scylla Memory carries durable project context forward across work and providers; it can inform an Agent, but Memory is context rather than execution authority.

Learn more: Team Memory →

12Can my Team enforce Agent Policy?

Yes. Managed Policy lets an organization establish rules that developers may tighten locally but may not weaken below the organization requirement. It governs supported Scylla-controlled operations; it does not replace Windows permissions, Git server permissions, database permissions, or cloud IAM.

Learn more: Managed Policy →

13How do Team invitations, seats, and desktop activation fit together?

They are separate steps: an invitation creates organization membership, a Team seat gives that member a Team entitlement, and the member then signs into the Scylla desktop to activate that entitlement on an installation. Accepting an invitation by itself does not assign a seat.

Learn more: Team Invitations → · Seat Assignment →

14Why did checkout succeed but my Team seat count not change immediately?

Returning from checkout is not the source of truth for Scylla licensing. Scylla uses verified server-side billing state, so the Team console must receive and display the updated subscription and purchased-seat quantity before those seats are available for assignment.

Learn more: Billing →

15What happens if Team Memory or another hosted Scylla service is unavailable?

A hosted-service problem should degrade that hosted capability rather than silently widen Agent authority. Local project work should continue where the rest of Scylla is available, and unavailable Memory should be reported as unavailable rather than incorrectly presented as an empty project history.

Learn more: Team Memory → · Troubleshooting →

16Where should I start when something is not working?

Start by identifying the failing layer: Scylla itself, account/provider authentication, Workspace or Knowledge Scope, Policy, Keyring/Connection, MCP, Memory, or the external service. That usually narrows the problem much faster than reconnecting or reconfiguring everything at once.

Learn more: Troubleshooting →

17How should I report a Scylla issue?

Include your Scylla and Windows versions, the affected provider or capability, the shortest reproduction steps, the expected result, and the sanitized error or screenshot. Never send passwords, API keys, private keys, raw .env files, session tokens, or secret-bearing connection strings in a support report.

Learn more: Contact / Issues →

Need help? Contact