Blades CLI
A CLI agent keeps memory in plain files, loads Markdown skills and runs recurring tasks through a persistent scheduler.
- 819 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The scheduler can execute shell commands and agent tasks without a person at the terminal.
Install
Manual install
git clone https://github.com/go-kratos/blades
cd blades/cmd/blades
go install .First installation method from the dedicated user CLI README.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The scheduler can execute shell commands and agent tasks without a person at the terminal.
This is third-party code. Review the repository files before installing.
What it does
Blades CLI runs a personal agent in the terminal and retains working context in files. Users can chat, execute a single request or schedule recurring work. Instructions and references live in a workspace where they can be read and edited directly. The scheduler supports shell commands, agent turns and notifications. A daemon executes schedules and optional communication channels. Interactive chat can run separately without keeping that daemon active.
Who it is for. Technical practitioners maintaining personal notes and recurring reports through an agent with inspectable memory.
Good fit when
- Every morning, you want a scheduled note review to use context from previous runs.
- For one project, you need a separate working directory and reference files.
- From the terminal, you want to complete a one-off task and exit.
- After a daemon restart, recurring work should retain its earlier history.
Not a fit when
- Background commands under your account are prohibited.
- A shared service needs tenant isolation and a ready permissions dashboard.
- The user expects a fully offline model without configuring a provider.
Example request
Prepare a daily review of notes in my workspace. Show the prompt and schedule before activating it.Limitations
Source installation requires Go 1.24 or newer; check it in the shell where you will build the CLI. Scheduled jobs run only while `blades daemon` is active, so the machine owner must maintain that process. History and memory stay local, including conversation text when `logConversation` is enabled; account for that in backups. Shell commands use the process environment; a workspace restriction setting does not replace operating-system isolation.
How to disable. Exit chat and stop the daemon. Inspect `blades cron list --all` to retain or remove jobs before restarting later.
FAQ
How do I initialize the workspace?
Run `blades init`. Existing files survive repeated initialization; inspect the model configuration afterward.
How do I start chatting?
Use `blades chat`. Specify `--session` to use a named session.
How are logs different from memory?
Runtime failures go to logs; context for later conversations goes to workspace memory. Start with runtime logs when diagnosing a failure.
How do I add a skill?
Place a directory containing `SKILL.md` under `~/.blades/skills/`. Review its commands and permissions first.
How do I check the installation?
Run `blades doctor` to inspect directories, required files and stale jobs. Resolve reported issues before enabling schedules.
Discussion
No questions or reviews yet.
Related
CLI toolsA tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
A local-first AI agent running on your own machine via llama.cpp that drives a browser, edits files, runs approved commands and remembers context across session
A scaffold of personas, rules and skills in plain Markdown: Maestro hands work to different models and spares your main plan
A bridge between local CLI agents: Codex, Cursor or Claude Code delegate work to Antigravity, Grok, Kimi and seven more.
A portable .agent folder with memory, skills and protocols that plugs into many coding agents and keeps its knowledge when you switch tools
One agent launches others in the background: Claude Code, Codex, Copilot CLI, Grok Build and Pi, with a model chosen per stage.
README in short
Memory separates lasting facts, daily notes and reference material. Optional Lark and Weixin channels connect to the background agent.