Reference
Configuration
Nothing is required. ~/.hum/config.toml is yours; <repo>/.hum/config.toml is the project's.
config.toml
[model]
name = "kimi-k3" # a route from /model, or provider/model
# reasoning_effort = "high" # low | medium | high; the server sends high unless you set this
# agents = "..." # route for sub-agents
# api_base = "https://..." # any OpenAI-compatible endpoint; with api_key_env, context_window
[[mcp]]
name = "jira"
url = "https://.../mcp" # or command = "npx", args = ["jira-mcp"]
[policy]
gate = "off" # off | approve
compact_at_tokens = 200000
[permissions]
allow = []
ask = []
deny = []
[hooks]
enabled = true
foreign = true # run Claude Code and Codex hook files
[sandbox]
mode = "workspace" # workspace | required | off
network = false
writable = []
env_pass = []
[sync]
world = true # send the working tree's history with each turn
[session]
delivery_question = true # ask whether you would merge an exchange that wrote files
[ui]
reasoning = true # stream the model's thinking
[notify]
desktop = true
Files
| Path | Holds |
|---|---|
~/.hum/config.toml |
Your settings |
<repo>/.hum/config.toml |
Project settings and rules |
~/.hum/hooks.json, <repo>/.hum/hooks.json |
Hooks |
~/.hum/skills/, .agents/skills/, .claude/skills/ |
Skills |
~/.hum/agents/, .hum/agents/ |
Sub-agent files |
~/.hum/credentials.json |
Sign-in |
~/.hum/calls/ |
Model calls on your own host |
~/.hum/blobs/ |
Images |
<repo>/.hum/tool-output/ |
Long tool output; swept after a week |
Environment variables
| Variable | Effect |
|---|---|
HUM_PLAIN=1 |
Text client without Node |
HUM_SANDBOX=off |
Sandbox off for one run |
HUM_ENV_PASS=A,B |
Pass variables to commands for one run |
HUM_HOOKS=0 |
Hooks off for one run |
HUM_DELIVERY_QUESTION=0 |
No delivery question for one run |
HUM_NOTIFY=0 |
No desktop notifications |
HUM_NO_AUTO_UPDATE=1 |
Notify, do not download |
HUM_NO_UPDATE_CHECK=1 |
No update box |
HUM_NO_AUTO_LOGIN=1 |
First run does not sign in |