chatdelta-cli-0.4.4 is not a library.

ChatDelta CLI

Query OpenAI, Gemini, and Claude in parallel and compare their responses — from a single command.

Installation

Requires a recent stable Rust toolchain:

cargo build --release

The binary lands at target/release/chatdelta.

Setup

Export whichever API keys you have — at least one is required:

export OPENAI_API_KEY=<key>       # or CHATGPT_API_KEY
export GEMINI_API_KEY=<key>
export ANTHROPIC_API_KEY=<key>    # or CLAUDE_API_KEY

Verify your configuration:

./chatdelta --doctor

Usage

./chatdelta "How do I implement a binary search in Rust?"

Common flags

Flag	Description
`--only gpt,gemini`	Query only the listed models
`--exclude claude`	Skip the listed models
`--system-prompt <text>`	Set a system prompt for all models
`--no-summary`	Skip the summary; show raw responses only
`--show-usage`	Print a token / latency table after responses
`--stream`	Stream tokens as they arrive (single-model; use with `--only`)
`--format text\|json\|markdown`	Output format (default: `text`)
`--log <path>`	Append the full exchange to a file
`--test`	Test API connectivity without sending a prompt
`--list-models`	Print available model names and exit

--show-usage

Appends a per-model token count and latency table after the response:

./chatdelta --show-usage "What is a monad?"

Model                  Tokens     Latency
──────────────────────────────────────────
ChatGPT                  1024      834ms
Gemini                    718      412ms
Claude                    892      612ms

--system-prompt

./chatdelta --system-prompt "You are a senior Rust engineer. Be concise." \
  "What are the trade-offs between Arc and Rc?"

Works in conversation mode too — see below.

--stream

Stream tokens from a single model as they arrive:

./chatdelta --stream --only claude "Explain monads in plain English."

If multiple models are selected, --stream falls back to parallel mode with a warning. When --stream and --show-usage are both set, streaming is skipped in favour of a metadata-bearing response so the usage table can be shown.

Conversation Mode

Start an interactive multi-turn session:

./chatdelta --conversation           # or -c
./chatdelta -c --system-prompt "You are a Socratic tutor."

Command	Action
`save`	Write history to the `--save-conversation` path
`clear`	Reset conversation history
`exit` / `quit`	End the session (auto-saves if `--save-conversation` is set)

Save and resume across runs:

./chatdelta -c --save-conversation session.json
./chatdelta -c --load-conversation session.json --save-conversation session.json

Debate Mode

Run a structured deliberation between two models on a proposition. A third model acts as moderator and produces a report covering the strongest point from each side, shared conclusions, unresolved disagreements, and factual claims worth verifying.

chatdelta debate \
  --model-a openai:gpt-4o \
  --model-b anthropic:claude-sonnet-4-6 \
  --moderator google:gemini-2.5-flash \
  --prompt "Microservices architecture improves long-term maintainability for most teams."

Models are specified as provider:model. Supported providers: openai, anthropic / claude, google / gemini. If --moderator is omitted, one is auto-selected from your available keys (Gemini → Claude → OpenAI).

Pipe from a file and export the full transcript:

cat proposition.txt | chatdelta debate \
  --model-a openai:gpt-4o \
  --model-b anthropic:claude-sonnet-4-6 \
  --rounds 2 \
  --export debate-output.md

deliberate is an alias for debate. Note: --stream, --system-prompt, --show-usage, and other top-level query flags do not apply to debate; use the debate-specific flags below.

Debate flags

Flag	Default	Description
`--model-a`	required	Model A: `provider:model`
`--model-b`	required	Model B: `provider:model`
`--moderator`	auto	Moderator: `provider:model`
`--prompt`	required	Proposition text (or `--prompt-file`, or pipe via stdin)
`--rounds`	`1`	Rebuttal pairs after the opening exchange
`--export`	—	Write transcript + moderator report to a markdown file
`--max-turn-chars`	`2000`	Character guideline per turn
`--quiet`	—	Suppress progress output

Development

cargo test
cargo test test_args_parsing    # run a single test

Tests use chatdelta v0.8.2 from crates.io. The mock feature is enabled in [dev-dependencies] so no live API keys are needed to run the suite.

Contributing

Contributions are welcome. Please read CONTRIBUTING.md for guidelines. This project follows our Code of Conduct.

License

MIT

chatdelta-cli 0.4.4