You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Multi-agent autonomous development system for Claude Code, OpenAI Codex CLI, and Google Gemini CLI
Transform your spec into a fully deployed, production-ready product with minimal human intervention. Built on 2025 research from OpenAI, Google DeepMind, and Anthropic.
Quick Start
# Pull the latest image
docker pull asklokesh/loki-mode:latest
# Show help
docker run --rm asklokesh/loki-mode
# Start autonomous mode with a spec
docker run -it \
-e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
-v $(pwd):/workspace \
asklokesh/loki-mode start prd.md
# With dashboard UI
docker run -it \
-e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
-p 57374:57374 \
-v $(pwd):/workspace \
asklokesh/loki-mode start --api prd.md
# Dashboard at http://localhost:57374
Image Details
Property
Value
Base
Ubuntu 24.04
User
loki (UID 1000, non-root)
Workdir
/workspace
CMD
loki help
Exposed Port
57374 (Dashboard/API)
Node.js
20 LTS
Python
3.x (for dashboard server)
GitHub CLI
v2.65.0
Usage Examples
# Interactive shell
docker run -it -v $(pwd):/workspace asklokesh/loki-mode bash
# Background autonomous mode
docker run -d \
-e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
-v $(pwd):/workspace \
asklokesh/loki-mode start --bg prd.md
# Quick single-task mode (max 3 iterations)
docker run -it \
-e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
-v $(pwd):/workspace \
asklokesh/loki-mode quick "add login page"# Check status
docker run -it -v $(pwd):/workspace asklokesh/loki-mode status
# Build a PRD interactively from templates
docker run -it -v $(pwd):/workspace asklokesh/loki-mode init
# Generate PRD from a GitHub issue
docker run -it \
-e GITHUB_TOKEN="$GITHUB_TOKEN" \
-v $(pwd):/workspace \
asklokesh/loki-mode issue https://github.com/org/repo/issues/42
Multi-Provider Support
# Claude (default) -- full feature support
docker run -it \
-e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
-v $(pwd):/workspace \
asklokesh/loki-mode start prd.md
# OpenAI Codex CLI -- degraded mode (sequential only)
docker run -it \
-e LOKI_PROVIDER=codex \
-e OPENAI_API_KEY="$OPENAI_API_KEY" \
-v $(pwd):/workspace \
asklokesh/loki-mode start prd.md
# Google Gemini CLI -- degraded mode (sequential only)
docker run -it \
-e LOKI_PROVIDER=gemini \
-e GOOGLE_API_KEY="$GOOGLE_API_KEY" \
-v $(pwd):/workspace \
asklokesh/loki-mode start prd.md
Volume Mounts
Host Path
Container Path
Mode
Purpose
Project dir
/workspace
rw
Source code and PRD files
~/.gitconfig
/home/loki/.gitconfig
ro
Git configuration
~/.ssh
/home/loki/.ssh
ro
Git SSH authentication
~/.config/gh
/home/loki/.config/gh
ro
GitHub CLI authentication
# Full setup with Git and GitHub access
docker run -it \
-e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
-e GITHUB_TOKEN="$GITHUB_TOKEN" \
-v $(pwd):/workspace \
-v ~/.gitconfig:/home/loki/.gitconfig:ro \
-v ~/.ssh:/home/loki/.ssh:ro \
-v ~/.config/gh:/home/loki/.config/gh:ro \
-p 57374:57374 \
asklokesh/loki-mode start --api prd.md
SSH Note: Prefer SSH agent forwarding over mounting private keys. Mount only known_hosts and public keys when possible.
Environment Variables
Credentials
Variable
Description
ANTHROPIC_API_KEY
Anthropic API key (required for Claude provider)
OPENAI_API_KEY
OpenAI API key (required for Codex provider)
GOOGLE_API_KEY
Google API key (required for Gemini provider)
GITHUB_TOKEN
GitHub personal access token
Core Configuration
Variable
Description
Default
LOKI_PROVIDER
AI provider: claude, codex, gemini
claude
LOKI_MAX_ITERATIONS
Max autonomous iteration cycles
1000
LOKI_MAX_RETRIES
Max retry attempts per iteration
50
LOKI_DASHBOARD
Enable dashboard server
true
LOKI_DASHBOARD_PORT
Dashboard/API port
57374
LOKI_BUDGET_LIMIT
Max USD spend before auto-pause (e.g. 50.00)
unset
LOKI_NOTIFICATIONS
Desktop notifications
false
Execution Control
Variable
Description
Default
LOKI_AUTONOMY_MODE
perpetual, checkpoint, or supervised
perpetual
LOKI_COMPLETION_PROMISE
Stop condition text (AI outputs this to halt)
unset
LOKI_PARALLEL_MODE
Enable git worktree parallelism
false
LOKI_MAX_PARALLEL_AGENTS
Limit concurrent sub-agents
10
LOKI_SKIP_MEMORY
Skip loading memory context
false
LOKI_SKIP_PREREQS
Skip prerequisite checks
false
Security (Enterprise)
Variable
Description
Default
LOKI_STAGED_AUTONOMY
Require approval before each action
false
LOKI_AUDIT_LOG
Enable audit logging
true
LOKI_ALLOWED_PATHS
Comma-separated writable paths
all
LOKI_BLOCKED_COMMANDS
Comma-separated blocked shell commands
rm -rf /
LOKI_SANDBOX_MODE
Run in Docker-in-Docker sandbox
false
SDLC Phases (all enabled by default, set to false to skip)
Start remote session (connect from phone/browser, Claude Pro/Max)
cleanup
Kill orphaned processes
version
Show version
help
Show help
Start Options
--provider NAME AI provider: claude (default), codex, gemini
--parallel Enable parallel mode with git worktrees
--bg, --background Run in background
--simple Force simple complexity tier
--complex Force complex complexity tier
--github Enable GitHub issue import
--no-dashboard Disable web dashboard
--sandbox Run in Docker sandbox
--skip-memory Skip loading memory context
--budget USD Set cost budget limit
--yes, -y Skip confirmation prompts