Multi-agent runtime
Give planning, coding, review, docs, and tests their own agents. Work moves between explicit roles, with an orchestrator coordinating the handoffs.
Explore the architectureBuilt for the
whole workflow
Plan, build, and pick up where you left off. Bring your own models and give every agent the tools and boundaries it needs.
Agents that work together
From the first plan to the final review, give every part of the job a clear role.
Goal mode
Set an objective with /goal. Bulb keeps working across iterations, checks its progress, and stops when the work is complete or needs attention.
Agent Skills
Add reusable instructions from a local folder or a Git repository. Bring compatible Claude Code skills into the same workflow.
Explore skillsApproval-aware execution
Choose how much freedom your agents have. In ask-first mode, review the plan and use /approve when you’re ready for work to begin.
Provider flexibility
Connect Anthropic, OpenAI-compatible providers, or local models through LM Studio and Ollama. Choose the models that suit your work.
Explore model setupExtended thinking
Use /thinking to adjust the reasoning effort for the active model. The runtime maps your choice to the connected provider.
A workflow that remembers
Keep the useful context, shape the tools, and make the next session easier to begin.
Persistent sessions
Return to the conversation, tasks, and agent activity you left behind. Resume saved work or compact a long thread without starting over.
Explore sessionsPersistent memory
Save preferences and project knowledge with /memory. Personal and project memories come along when you start the next session.
Custom slash commands
Turn a prompt you use often into a command. Keep it in your personal setup or share it with a repository.
Explore custom commandsRuntime hooks
Add your own checks around tool use, permission requests, and session startup. Keep project-specific rules close to the work.
Explore hooksWeb tools
Search the web, read pages, and download resources from the same session. Network access follows the configured approval flow.
Explore web toolsClipboard & attachments
Bring images and files into the conversation, copy a response, or select text with dedicated keyboard shortcuts.
Explore attachmentsConnections with clear boundaries
Choose what agents can reach and how you want to work with them.
Encrypted key storage
Provider API keys are encrypted with AES-GCM. Configuration, trusted paths, and saved sessions stay in the local harness storage.
Explore configurationOS sandboxing
Set filesystem and network policies for shell commands. Bulb uses the available operating-system controls, with platform-specific capabilities documented in the guide.
Explore sandboxingRemote, Telegram & ACP
Control a session remotely, continue from Telegram, or connect an ACP-capable editor. Use the same runtime from the tools around you.
Explore remote connectionsMake room for your next idea
Open a project and put Bulb to work.
Built on the open-source Spettro harness. Explore the implementation and platform details in the Bulb guides.