agntk

AI agent you install and point at problems. Zero config. Works immediately.
No API keys required — a free tier is included. Bring your own key or install Ollama for unlimited local use.
---
Quick Start
That's it. No config files. No API key setup (unless you want to). It reads your files, runs commands, browses the web, spawns sub-agents, and remembers what it learns.
After any one-shot prompt, you can type follow-up messages directly — the agent stays in the terminal and remembers the conversation. Use -q to disable this and exit immediately.
---
What It Does
Out of the box, every agent has 20+ built-in tools:
- **Files** — read, write, edit, create, glob, grep across your workspace
- **Code** — AST-aware search and replace via ast-grep
- **Shell** — run commands, manage background processes
- **Browser** — navigate, extract, interact with web pages
- **Planning** — break down complex tasks, deep reasoning for hard problems
- **Memory** — remember facts, recall context, build knowledge across sessions
- **Sub-agents** — spawn specialized agents for parallel work, with live activity streaming
- **Skills** — auto-discover SKILL.md files for project-specific capabilities
- **Governance** — three-tier behavioral framework (CORE identity > human rules > agent instincts)
- **Growth loop** — agents log events, synthesize daily journals, and create learned instincts
---
Zero-Config Provider Cascade
agntk auto-detects the best available AI provider:
| Priority | Provider | How it's detected |
| -------- | ---------------- | ---------------------------------------------------------------------------- |
| 1 | **Your API key** | OPENROUTER_API_KEY, OPENAI_API_KEY, or CEREBRAS_API_KEY in env |
| 2 | **Ollama** | Auto-detected at localhost:11434 — picks the right model for your hardware |
| 3 | **Free tier** | Built-in, no setup — backed by Cerebras (rate-limited) |
---
CLI Reference
Options
| Flag | Short | Description |
| ---------------- | ----- | ------------------------------------------- |
| --name | -n | Agent name (enables persistent memory) |
| --instructions | | Custom system prompt |
| --interactive | -i | Interactive REPL mode |
| --workspace | | Workspace root (default: current directory) |
| --max-steps | | Max tool-loop steps (default: 25) |
| --verbose | | Show full tool args and output |
| --quiet | -q | Text output only (no follow-up, for piping) |
| --version | -v | Show version |
| --help | -h | Show help |
Agent Management
| Command | Description |
| --------------- | --------------------------------------------------------------------------- |
| list | Show all agents with status (running/idle) and last active time |
| info | Agent details — memory files, workspace, sub-agents, token usage, disk size |
| delete | Delete an agent's state directory (with confirmation) |
| stop | Send SIGTERM to a running agent; SIGKILL if it doesn't exit |
| clean | Interactive picker to bulk-delete idle agents |
| completions | Output shell completion script (bash, zsh, fish) |
| install | Validate and install a capability (rule, instinct, skill) into the harness |
| uninstall | Remove an installed capability file |
| evaluate | Validate a capability file without installing it |
Interactive REPL
In REPL mode (-i or follow-up after one-shot), press TAB after / for autocomplete:
| Command | Description |
| ---------- | -------------------------- |
| /help | Show available commands |