henosis-cli 0.6.32

henosis-cli — interactive CLI for the Henosis multi-provider streaming chat backend, with optional local tools.

Henosis CLI

the BEST terminal agent designed for uncompromising performance

Henosis CLI is a streamlined, professional terminal client for the Henosis multi-provider streaming chat backend. It supports OpenAI, Gemini, Anthropic, xAI (Grok), DeepSeek, and Moonshot Kimi via the Henosis server, and includes optional client-executed file/shell tools with approvals and sandboxing.

Key features

• Interactive chat over SSE with usage/cost summaries
• Model picker and per-turn controls (tools on/off, control level, reasoning effort)
• Client approvals at Level 2 for write/exec operations (approve once/session/always)
• Agent scope (safe host directory) when enabling tools in host mode
• Optional web search controls for OpenAI models (domain allow-list, include sources, location hints)
• Saves conversations to server threads and commits usage for billing where enabled

Troubleshooting

• If a provider streams deltas but never sends a final message.completed, the CLI now prints a Diagnostics block with:
  • stream stats (events/deltas/bytes), last events tail, and response x-* headers
  • It also writes the same data to logs/session-*.jsonl as event=diagnostics.no_completed
• Make the tail longer with HENOSIS_CLI_SSE_TAIL (default 40). Example: HENOSIS_CLI_SSE_TAIL=80 henosis-cli
• For raw debugging: use --debug-sse and/or --debug-req
• Minimal harness: python test_stream.py (override with HENOSIS_TEST_URL, HENOSIS_TEST_MODEL, HENOSIS_TEST_PROMPT)

Install

• pip: pip install henosis-cli
• pipx (recommended): pipx install henosis-cli

Quick start

• Run the CLI: henosis-cli
• Default server: https://henosis.us/api_v2 (override with HENOSIS_SERVER or --server)
• Dev server: henosis-cli --dev (uses HENOSIS_DEV_SERVER or http://127.0.0.1:8000)
• Authenticate when prompted. Tools are enabled by default; use /model to pick a model and /tools off if you want them disabled.

Common commands

• /menu or /settings: Open settings menu
• /model: Open settings model picker (or '/model ' to set directly)
• /tools on|off|default: Toggle per-request tool availability
• /fs workspace|host|default: Set filesystem scope (workspace = sandbox; host = Agent scope)
• /hostbase
• /level 1|2|3: Control level (1 read-only; 2 write/exec with approval; 3 no approvals)
• /map on|off: Inject CODEBASE_MAP.md into your first message
• /websearch on|off|domains|sources|location: Configure OpenAI web search options
• /title : Name the current chat thread
• /clear: Reset chat history
• /login, /logout, /whoami: Auth helpers

Configuration

• Server base URL
  • Env: HENOSIS_SERVER (default https://henosis.us/api_v2)
  • Flag: --server https://your-server
  • Dev shortcut: --dev (env HENOSIS_DEV_SERVER or http://127.0.0.1:8000)
• Optional Agent Mode (developer WebSocket bridge): --agent-mode

Local tools and sandboxing (optional)

• The CLI can execute a safe subset of tools locally when the server requests client-side execution.
• Tools include read_file, write_file, append_file, list_dir, apply_patch, run_command, spawn_command, process_list, process_status, process_output, wait_process, terminate_process, spawn_terminal, terminal_list, terminal_snapshot, terminal_input, terminal_resize, terminal_close.
• At Level 2, destructive tools and command executions prompt for approval (once/session/always).
• Workspace root: by default, the workspace scope is the current working directory at the moment you launch the CLI. No dedicated per-terminal sandbox is created unless you override it.
• Override root: set --workspace-dir /path/to/root (or HENOSIS_WORKSPACE_DIR) to operate in a different directory for the session.
• Host scope can be constrained to an Agent scope directory (set via /hostbase) when fs=host.

Notes

• Requires Python 3.9+
• The CLI ships with rich and prompt_toolkit for a nicer UI by default.
• The reusable local tools library is available as a module (henosis_cli_tools).

Anthropic-only context handoff tool (server owners)

• The server can expose a lightweight, provider-scoped function tool to Anthropic models only that lets the model signal a handoff to the next turn and carry an optional JSON payload.
• Enable by setting on the API server (not the CLI):
  • ANTHROPIC_ENABLE_TONEXT_TOOL=true
  • Optional: ANTHROPIC_TONEXT_TOOL_NAME=context (default) or to_next (legacy)
  • Optional: ANTHROPIC_TONEXT_ATTACH_BETA=true and ANTHROPIC_TONEXT_BETA_HEADER=context-management-2025-06-27
• When enabled, clients will see normal tool.call/tool.result SSE events. The tool.result includes {"to_next": true} and echoes the provided payload.
• No client-side tool execution is required; the server handles this tool inline.

Support

• Email: [e-mail]

Build and publish (maintainers)

• Bump version in pyproject.toml
• Build: python -m pip install build twine && python -m build
• Upload to PyPI: python -m twine upload dist/*
• Or to TestPyPI: python -m twine upload --repository testpypi dist/*