Loki Mode CLI
An autonomous spec-driven builder: takes a PRD or issue, writes the code and refuses to call it done without verification
- 1,111 ★ on the repository
- Last commit
Broad access
Why this level
- The agent runs commands, writes code and commits autonomously
- Skill mode requires skipping permission prompts
- Consumes paid model tokens
Install
- Works in
Manual install
npm install -g loki-modeAlternatives: bun install -g loki-mode (recommended by the author) or brew tap asklokesh/tap && brew install loki-mode.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs commands, writes code and commits autonomously
- Skill mode requires skipping permission prompts
- Consumes paid model tokens
This is third-party code. Review the repository files before installing.
What it does
Loki Mode takes a spec as a PRD, GitHub issue, OpenAPI document or a one-line idea and runs planning, implementation, review and verification on its own. Work is not accepted until eight quality gates pass: an empty diff, failing tests, an app that does not boot or a leaked secret all block completion. Each run writes a receipt to .loki/proofs/ that separates facts (diff, test and build exit codes) from model judgments, and loki proof verify re-checks it. For existing repositories a modernize mode starts with a read-only assessment of the codebase.
Who it is for. For developers and founders who want to hand an agent a whole build and get a result they can verify.
Good fit when
- You have a PRD or issue and want it turned into working code with tests
- You need a prototype from a short brief without steering the agent by hand
- You want to assess a legacy repository before modernization without changing it
Not a fit when
- You need small edits under close control
- You cannot let an agent run commands without approval
- Your token budget is tight
Example request
Run Loki Mode on prd.md and build the app with testsLimitations
Requires ANTHROPIC_API_KEY or another model provider (Claude Code, Codex, Cline, Aider). The autonomous skill mode requires the --dangerously-skip-permissions flag. Licensed under BUSL-1.1 rather than a classic open source license; commercial editions are sold under the Autonomi brand. Long builds consume significant tokens; quickstart shows a cost estimate before starting.
How to disable. Remove the CLI with npm uninstall -g loki-mode or brew uninstall loki-mode and uninstall the Claude Code plugin with claude plugin uninstall. A running session stops when a .loki/STOP file appears in the project.
FAQ
Can I try it without installing or a key?
Yes, npx loki-mode tour prints a sample receipt from a past build with no network or spend.
What does the Claude Code plugin add?
Three slash commands and the Loki MCP server with memory, a task queue and code search. The CLI itself must be installed separately.
Discussion
No questions or reviews yet.
Related
CLI toolsA Ralph loop for long tasks: an agent in a Docker sandbox works through a PRD task list, checks each task with tests and commits until the list is done
A Rust CLI that runs development as a team of roles on top of an installed Claude Code, Codex or OpenCode: plan, checks and a final report
A completion-discipline skill for large agent tasks: write the acceptance ledger with runnable checks first, then do the work, then reverify against evidence
Local coding agent with a terminal UI: it plans, drafts several candidates, tests them in a sandbox and repairs, so compact open models can handle code
A CLI that detects your project's stack and installs matching agent skills from a vetted registry in one command
Skills for cross-provider work between Claude Code and Codex: one agent drafts the plan, the other reviews it, then a build and an independent code inspection f
README in short
The README opens with installation via bun, npm, Homebrew or Docker and the loki quickstart command, which asks a few questions, shows a cost estimate and builds the project. It then covers a non-interactive CI mode, template selection, JSON plan previews and a modernize mode for existing repositories. A dedicated section explains verification receipts and states plainly what they do not guarantee. The project works with Claude, Codex, Cline and Aider.