Using Codebolt
Reference documentation for every feature. Jump to what you need — this isn't meant to be read cover-to-cover.
For step-by-step walkthroughs see Guides & Tutorials. For writing code against Codebolt see Build on Codebolt.
Start Here
New to Codebolt? Start with what it is, get it installed, and run your first agent.
The core idea, what makes it different, and what you can do with it.
→QSQuickstartInstall, sign in, open a project, and run your first agent in ~10 minutes.
→InstallInstallation & SetupDesktop app, CLI, onboarding wizard, and first-run checklist.
→Workspace
The interfaces you interact with day-to-day — where you write, run commands, and talk to agents.
Conversations with agents, @-mentions, checkpoints, and inline editing.
→CodeCode Editor FeaturesCode editor, terminal, git panel, and preview browser.
→AppApplication FeaturesAuthentication, teams, billing, and application layout.
→ObsCode ObservabilityCodemap, narrative graph, and structural execution views for understanding project architecture and run flow.
→Agents & Tools
Everything to do with running AI agents — what they are, where they run, what tools they use, what they remember, and how to coordinate many of them at once.
Install, run, debug, and manage AI agents and portfolios.
→EnvEnvironmentsRun agents locally, in Docker, E2B, Daytona, or on a remote server.
→ToolsAgent ToolsBuilt-in tool families and panel tools available to every agent.
→MemMemory & ContextHow agents remember and retrieve information across sessions.
→MultiMulti-Agent UsageRun swarms and flows, review outputs, trace execution.
→Agent Extensions
Modular units that extend what agents can do — reusable capabilities, slash-command skills, parallel action blocks, the runtimes that execute them, and MCP servers for external tool access.
Versioned, reusable bundles of agent behaviour with typed inputs and outputs.
→SkillSkillsSlash-command invocable capabilities installable from the marketplace.
→ABAction BlocksLightweight code units that run as side executions parallel to an agent.
→ExecExecutorsLanguage runtimes (Node, Python, shell) that run capabilities.
→Eval & Optimization
Measure agent quality with configurable scoring, then automatically improve agents through an AI-driven optimization loop.
Define test tasks with instructions, environments, and weighted evaluators.
→ScoreEvaluatorsExpected output, script, agent judge, and deliberation scoring mechanisms.
→RunRunning EvalsCreate runs, monitor results in real time, and compare agents on a leaderboard.
→OptOptimizationAuto-improve agents across iterations using greedy, best-of-N, or annealing strategies.
→Multi-Agent Coordination
Tools for coordinating work across multiple agents — distributing tasks, messaging, reaching consensus, and managing swarms and orchestrators.
Work queue agents bid on, lock, split, and deposit pheromones to coordinate around.
→MailMail & InboxThreaded asynchronous messaging between agents and users.
→DelibAgent DeliberationVoting, feedback, and consensus mechanisms for shared decisions.
→SwarmSwarm ManagementCreate and manage named groups of agents working toward a shared goal.
→OrchOrchestrator ManagementRoute tasks across agents via an active orchestrator layer.
→Planning
Structured tools for managing work from high-level strategy down to individual tasks. All planning data lives inside your project as plain files you can commit and diff.
Features and phases, with an Ideation tab for reviewing suggestions.
→SpecsSpecsTechnical specification documents with rich Markdown and inline comments.
→TasksTasksIndividual work items with kanban view, subtask hierarchy, and agent message log.
→ActionAction PlanStructured execution plans with tasks, dependencies, parallel groups, and conditionals.
→UIUI FlowWireframes and mockups on a freehand canvas.
→ReqRequirement PlanAggregate documents linking Specs, UI Flows, and Action Plans.
→Integrations
Connect agents to external platforms — chat apps, project tools, and any custom service — through the Routing Gateway and a plugin-based channel system.
Central message router — rules, thread strategies, and activity log.
→ChatChat PlatformsConnect agents to Telegram, Slack, Discord, Teams, and WhatsApp via plugins.
→ToolsProject ToolsIntegrate with Linear, GitHub, Jira, and Notion via MCP servers or webhooks.
→PluginBuilding a Channel PluginConnect any custom platform to the gateway with a WebSocket plugin.
→Automation
Mechanisms that trigger agents automatically — without a user typing a message. Route incoming HTTP requests, schedule timed events, react to file and git changes, and trace everything through the event log.
Incoming HTTP endpoints that route external requests to agents.
→CalCalendar EventsSchedule one-off and recurring events that notify agents at a set time.
→HookHooksReact to file saves, git commits, and agent lifecycle events automatically.
→LogEvent LogAppend-only audit trail of every event — lives under Memory & Context as the audit layer.
→Custom Applications
Build focused agent-powered products on top of Codebolt by combining a custom UI, domain-specific agents, and the right automation and integration surface.
Cloud
The hosted portal at portal.codebolt.ai — run and scale agents in cloud sandboxes, bring your own E2B or Daytona key, and publish to the marketplace.
What Codebolt Cloud gives you, when to use it, and when local is enough.
→ChatRemote ChatChat against an agent running in a cloud sandbox, optionally bound to a GitHub repo.
→RTRuntimes & ProvidersManage cloud sandbox instances and bring your own E2B or Daytona credentials.
→Configuration
Connect external services, set guardrails, and manage your account. Usually a one-time setup per project or workspace.
If you want to build your own product-specific UI or plugin-backed application on top of Codebolt, see Build on Codebolt → Custom Agentic Applications.
LLM providers, local models, git, browser control, chat widget, and plugin debug.
→GuardGuardrails & SettingsSafety rules, routing gateway, app and project config.
→Operations
Tools for monitoring what's happening — track running jobs, debug agent behaviour, and diagnose errors.