Octomind Documentation

Use these guides to install, configure, and operate Octomind's CLI agent runtime, or to contribute to its Rust code.

Use these guides to install, configure, and operate Octomind's CLI agent runtime, or to contribute to its Rust code. Start with a terminal session, then follow the guides for automation, integrations, and development.

Start Here

Install the binary and sign in to an Octomind account to use the shipped OctoHub model default:

bash
curl -fsSL https://octomind.run/install.sh | bash
octomind login
octomind

octomind login asks you to confirm a code in the browser and CLI, then stores the hub key in the user-scope config/.env and account credentials in config/auth.json. It does not change your model selection. New configs use assistant:concierge as the default tag and octohub:auto as the model name. See Installation and Quickstart for the complete setup.

Configure

Inspect and validate your configuration before adding roles or tools:

bash
octomind config --show
octomind config --validate

The default config is under ~/.local/share/octomind/config/ on Linux/macOS and %LOCALAPPDATA%/octomind/config/ on Windows. OCTOMIND_DATA_DIR relocates Octomind state; for example:

bash
OCTOMIND_DATA_DIR="$PWD/.octomind-data" octomind config --show

Use Configuration for file merging and overrides, or AI Providers to supply provider credentials yourself.

Operate

Create or resume a named interactive session:

bash
octomind run --name project-review
# After exiting, continue the same session:
octomind run --resume project-review

Inside a session, discover commands and inspect active tools:

text
/help
/mcp
/exit

For automation, pipe the prompt through stdin and choose plain or jsonl output:

bash
printf '%s\n' 'Summarize the responsibilities of an MCP server in three sentences.' | octomind run --format plain

run takes a role/tag positional argument, not a message argument. See Daemon & Hooks for persistent workers and Workflows for multi-step runs.

Usage Guide

DocumentDescription
InstallationRecommended setup, alternative installs, and shell completions
QuickstartFirst session, common commands, and non-interactive use
ConfigurationConfig locations, merging, models, roles, and MCP servers
ProvidersOctoHub, provider credentials, model selection, and local CLI backends
SessionsInteractive sessions, persistence, and multimodal input
RolesRoles, prompts, permissions, and tool access
MCP ToolsBuilt-in tools and runtime tool management
CompressionAutomatic context compression
WorkflowsMulti-step AI processing workflows
Commands & LayersCustom commands, layers, agents, and prompts
Structured OutputJSON Schema output for automation
Editor IntegrationNeovim, Zed, and JetBrains setup
LearningCross-session adaptive learning
SupervisorCompletion checks, planning, condensation, and learning control
SkillsAuto-activating skills and validators
Token EfficiencyContext and capability efficiency
Local ToolsProject-local scripts exposed as MCP tools
GuardrailsDeterministic project policies and hooks

Integration Guide

DocumentDescription
WebSocket ServerRemote sessions over WebSocket
ACP ProtocolAgent Client Protocol integration
Daemon & HooksLong-running sessions and webhook listeners
Tap SystemAgent, skill, capability, and workflow registries

Use Cases

DocumentDescription
CI/CD Code ReviewAutomated review with structured output
Event-Driven AgentDaemon sessions driven by webhooks
Custom WorkflowMulti-stage development workflows
Web DashboardEmbedding sessions through WebSocket
Multi-Agent DelegationDelegating work to specialized agents
Dynamic MCP ServersRuntime tool-server configuration
Scheduled TasksTimed messages and recurring work
Long-Running DevelopmentNamed sessions and resume workflows
Custom HooksScript-backed webhook integration
Safe Agent: Sandbox & GuardrailsBlock destructive commands and verify the agent’s work
Predictable AI SpendSpending checkpoints, per-role models, and usage reports
Make “Done” Mean DoneSupervisor gate, plan checks, and a test validator
Project Conventions via SkillsAuto-activating skills and convention validators
Learn from Past SessionsCarry corrections into the next session
Project Scripts as ToolsLet the agent run your checks through local tools
Build & Share a SpecialistCreate, publish, and install a tap agent
Understand a CodebaseSource-backed onboarding to an unfamiliar repo
Private Code, Local ModelsRoute private roles to Ollama, others to the cloud
Second Opinion via Fan-outParallel branches and a judge step
Research with Dynamic Fan-outPlanner-driven parallel investigations
Implement–Review–Fix LoopBounded graph workflow until review passes
Release Notes from Git HistoryPlain drafts and schema-validated release data

Development Guide

DocumentDescription
Building from SourceRust setup and development builds
ArchitectureSource modules and internal flows
MCP Server DevelopmentBuilding MCP servers for Octomind
Learning BenchmarkRetrieval and consolidation contract benchmark

Common Questions

If the shell cannot find the binary after the default installation, add the installer directory to your current shell's path:

bash
export PATH="$HOME/.local/bin:$PATH"
octomind --version

If login cannot open a local browser, print the URL instead:

bash
octomind login --no-browser

If a piped command hangs, ensure its producer closes stdin; Octomind reads the complete prompt before starting. For configuration and credential errors, start with octomind config --validate above and the guides below.

Troubleshooting and Reference

DocumentDescription
Common IssuesInstallation, configuration, provider, and session problems
Migration GuideUpgrading legacy configurations
CLI ReferenceCLI subcommands and flags
Session CommandsInteractive slash commands
Config ReferenceConfiguration fields and defaults
Environment VariablesCredentials, overrides, and runtime variables

See also