01 One Policy
Files · Git · Database · Shell · MCP · Skills
AI Development Workbench
Let agents work. Keep the authority.
Scylla is a development workbench for serious AI-assisted engineering. Use OpenAI, Claude, and Cursor ACP with the files, terminal, MCP tools, Skills, project context, and connections they need while Scylla keeps policy, credentials, and consequential authority under your control.
OpenAI Codex · Claude Code · Cursor ACP
02 One Broker
Protected capability without reusable secrets
03 Project Memory
Plans · Blueprints · Handoffs · Decisions
04 Your Agents
OpenAI · Claude · Cursor ACP
Works With
Bring the agents you already use. Give them one controlled place to work.
Scylla does not sell a model. Provider authentication and model availability remain provider-owned. Scylla supplies project context, operational capabilities, and an authority boundary around OpenAI Codex, Claude Code, and Cursor ACP.
01Paired
OpenAI Codex
02Paired
Claude Code
03Paired
Cursor ACP
Provider-owned authentication
Use your provider account. Authenticate through the supported provider path. Scylla licensing remains separate from provider access.
Use supported API configuration. Where a supported path uses protected configuration, Scylla keeps reusable values out of ordinary agent context.
Provider access and Scylla entitlement remain independent.
Bring the intelligence you prefer
Claude Code
claude-sonnet-4-6
claude-opus-4-1
Codex
gpt-5.5-codex
Use the models available through your configured providers. Scylla supplies the workspace, context, capabilities, and authority boundary.
01Control
Prompts express intent. Policy defines authority.
Supported Files, Git, database, shell / SSH, MCP, and operational capabilities resolve through one ALLOW / ASK / BLOCK model.
An imperfect prompt does not automatically become permission for a destructive production operation. Teams can enforce a baseline that developers may tighten but never weaken.
Explore Policy →OperationAllowAskBlock
Git status
Git push
Force push
Project read
Project delete
Outside-project
DB SELECT
DB UPDATE
DB TRUNCATE
02Brokered Authority
Use the credential without giving the credential away.
For supported brokered operations, a saved connection represents bounded capability. Scylla evaluates policy, resolves the protected credential internally, executes the approved operation, and returns the result.
The operation gets the credential. The agent does not.
How the Broker Works →
03Continuity
The project remembers even when the agent changes.
Scylla Memory keeps Plans, Blueprints, Handoffs, Knowledge, reports, and indexed decisions available across chats, providers, developers, and time.
Switch the model. Switch the developer. Keep the project context.
Explore Scylla Memory →
Capability
More control. More useful agents.
Build
Files · Editor · Git · Terminal · Tasks
Connect
MCP · Connections · supported SSH / Database
Operate
Skills · Environments · Cross-project runs
Remember
Knowledge · Memory · Plans · Blueprints · Handoffs
Understand
Estate · Maintenance · Attention · Release Sets
Solo / Team
More capability does not require less control.
Your machine. Your rules.
Set personal Global Policy, keep a local Keyring and Memory, and use project-scoped connections with the same brokered authority model.
- Personal Global Policy
- Local Keyring
- Local Scylla Memory
- Project-scoped connections
One authority ceiling across the team.
Managed Policy, Team Access, and Team Memory let developers work inside a shared authority ceiling they may tighten locally.
- Managed Global Policy
- Seat-based Team licensing
- Team Memory
- Member and installation management
Give agents more capability without giving away the machine.
OpenAI · Claude · Cursor ACP — with Policy · Broker · Memory · Skills · Connections.