OpenAI builds and deploys advanced AI models like GPT-4o for autonomous agents and workflows.
Caelis
0Popularity
About Caelis
Caelis is a local-first workspace designed for models and ACP agents to collaborate on tasks. Participants maintain their own conversations while exchanging progress and findings through messages, enabling coordinated work on complex problems. The tool allows users to bind specific models or agents to custom roles such as Guardian, Reviewer, or Memory Steward, configuring their behavior separately. Sessions, configuration, and credentials are stored locally under ~/.caelis, with the option to resume earlier work. Caelis supports both provider-based models and ACP agents, enabling tool approvals and local command execution under configured sandbox policies. Operating-system isolation limits filesystem access for local commands, and users can switch between auto-review and manual approval modes for tool requests. The source code is available under the Apache-2.0 license.
Key features
- Local collaboration workspace for models and agents
- Custom role binding (Guardian, Reviewer, Memory Steward)
- Auto-review and manual approval modes for tool requests
- Operating-system sandboxing for local commands
- Session resumption for continuing earlier work
- Provider and ACP agent connection management
- Local credential and configuration storage
- Cross-platform installation via CLI or npm
Use cases
- Coordinating multi-model workflows on complex tasks
- Debugging and reviewing code with sandboxed local commands
- Managing agent-based workflows with role-specific configurations
Pros
- Local-first architecture with offline session storage
- Supports both provider-based models and ACP agents
- Configurable sandboxing for local command execution
- Customizable roles and approval policies
- Open-source under Apache-2.0 license
Cons
- No explicit support for Windows GUI installation
- Limited to local execution with no cloud sync mentioned
- Requires manual setup for provider connections
Frequently asked questions about Caelis
What is Caelis and who is it designed for?
Caelis is a local-first workspace designed for models and ACP agents to collaborate on tasks. It is intended for users who need to coordinate multiple AI systems or agents to work together on complex problems.
How do agents collaborate in Caelis?
Participants maintain their own conversations while exchanging progress and findings through messages. Users can coordinate work from the main conversation or open a participant pane to follow up directly.
What roles can be assigned to models or agents in Caelis?
Users can bind models or ACP agents to custom roles such as Guardian, Reviewer, or Memory Steward, and configure their behavior separately for specialized tasks.
What sign-in and provider connection methods are supported?
Users can sign in with ChatGPT, Codex, or Grok, configure API-key or local model providers, or connect an installed ACP agent using the /connect command.
Where is data stored in Caelis?
Sessions, configuration, and Caelis-managed credentials are stored locally under ~/.caelis by default. Users can resume earlier work using the /resume command.
How does Caelis handle tool approvals and security?
Tool requests follow an approval mode, with local commands running under a configured sandbox policy. Users can switch between auto-review and manual approval modes for tool requests.
