chamber-cli 0.3.0

Private expert panels in your terminal

Chamber CLI

Private expert panels in your terminal.

Chamber CLI runs multi-agent AI discussions entirely on your machine using local models. No cloud, no accounts, no telemetry.

Install

pipx install chamber-cli

Or with pip:

pip install chamber-cli

Prerequisites

Install Ollama and pull a model:

ollama pull llama3.1

Usage

# Interactive REPL (default)
chamber

# Start with a topic
chamber "What are the legal risks of publishing leaked documents?"

# One-shot mode (no REPL)
chamber "Compare Signal vs Session for whistleblowers" --one-shot

# Custom settings
chamber --model mistral --agents 4 --rounds 5

REPL Commands

Command	Action
/follow <text>	Inject a follow-up into the next round
/rounds <n>	Set max rounds (1-5)
/agents	List current panel
/export	Export as markdown to stdout
/export --encrypt	Export with passphrase encryption
/save <path>	Save export to file
/new	Clear session, new topic
/status	Show provider, model, stats
/quit	Exit

Providers

Provider	Type	Default Port
Ollama	Local	localhost:11434
LM Studio	Local	localhost:1234

Privacy & Security

What Chamber CLI does

• Zero telemetry — no analytics, no tracking, no phone-home
• Zero disk writes — sessions exist only in memory, destroyed on exit
• No config files — settings via flags and environment variables only
• No shell history — REPL input is not written to any history file
• Encrypted export — AES-256-GCM with passphrase when you choose to save
• Minimal dependencies — small, auditable dependency tree
• Open source — read every line of code yourself

What Chamber CLI does NOT do

• No account system
• No API key storage
• No crash reporting
• No auto-updates
• No fingerprinting

Threat Model

Chamber CLI is designed to keep your AI discussions private from third parties. Here is what it protects against and what it does not:

Protected:

• Third-party data collection (no network calls in local mode)
• Persistent data leakage (nothing written to disk by default)
• Session recovery after exit (memory is freed)

Not protected:

• A compromised operating system or keylogger
• Memory forensics on a running machine
• Shell history in one-shot mode (topics appear in shell history — use REPL mode or pipe from stdin: echo "topic" | chamber --one-shot)

Remote mode (future):

• Remote providers send data over the network
• Use --proxy socks5://localhost:9050 for Tor routing
• Verify the provider's privacy policy independently

Environment Variables

Variable	Default	Description
CHAMBER_PROVIDER	ollama	Provider name
CHAMBER_MODEL	(provider default)	Model name
CHAMBER_AGENTS	3	Number of agents
CHAMBER_ROUNDS	3	Max rounds
CHAMBER_OLLAMA_URL	http://localhost:11434	Ollama server URL
CHAMBER_LMSTUDIO_URL	http://localhost:1234	LM Studio server URL
CHAMBER_PROXY	(none)	SOCKS5 proxy URL

Docker

docker run --rm -it --network host ghcr.io/reeseysan/chamber-cli

License

MIT