TheDocumentation Index
Fetch the complete documentation index at: https://ahvn.top/llms.txt
Use this file to discover all available pages before exploring further.
hb command is the shallow command-line layer over the same HeavenBase APIs used from Python. It initializes global config, browses registered workspaces, edits config, and calls configured LLM presets.
Setup and paths
hb setup initializes the global HeavenBase config store and registers the default workspace. It does not create a local .heavenbase/ folder.
hb pj resolves HeavenBase path aliases such as %/ for package state and &/ for resources.
Workspaces
hb ws manages the global workspace registry.
list|ls, activate|act|use, deactivate|deact, set|add, and unset|remove|rm|del|delete.
Workspace presets mirror the Python API hb.HeavenBase("name", preset="debug"). Use debug for no-Docker demos, local-lts for the local service stack, and web-lts for managed deployments.
Config
hb config edits HeavenBase config through ConfigManager. The short alias is hb cfg.
--scope to target a config layer, --version for retained history reads, and --json for JSON output or JSON value parsing.
Aliases: hb cfg ls / show, hb cfg rm / remove / del / delete, hb cfg cp, and hb cfg hist.
LLM
hb llm calls the LLM config system. Presets, providers, models, and gateways resolve the same way as hb.LLM(...) in Python.
--preset / -p, --model / -m, --provider / -b, --gateway / -g, --cache / --no-cache, and --verbose / -v on call commands. chat, embed, and session also accept repeated --input files. chat and session accept repeated --mcp values for MCP URLs or registered Toolkit refs such as quickstart.math-tools:-1. chat --max-steps caps MCP tool loops and defaults to 20 assistant steps.
The canonical Toolkit ref form is namespace.toolkit:version. Negative versions select from newest to oldest: -1 is latest, -2 is second-most latest, etc.
hb llm verbose shows the resolved non-secret LLM args. resolve remains an alias and appears in help as verbose|resolve. hb llm list|ls lists presets, providers, models, gateways, or all.
Reasoning chunks from models that emit think content are printed in grey inside <think> and </think> before normal answer text. MCP tool loops print each assistant iteration as STEPS: 001 / 020.
Interactive hb llm session supports /help, /save, /load, /clear, /regen, /back, /tools, /mcp, and /exit with short forms such as /s, /l, /r, and /b.

