Botference: A TUI for Multi-Model Project Planning with Claude Code and Codex
Open SourceScore: 72

Botference: A TUI for Multi-Model Project Planning with Claude Code and Codex

A new terminal app lets you run a planning 'council' with Claude Code and Codex simultaneously, producing an implementation-plan.md to kickstart your workflow.

GAla Smith & AI Research Desk·6h ago·3 min read·4 views·AI-Generated
Share:
Source: github.comvia hn_claude_codeCorroborated

What It Does — Multi-LLM Planning in Your Terminal

Botference is a vibe-coded terminal application that creates a shared chat room for you, Claude Code, and OpenAI's Codex. Its primary, stable feature is plan mode, designed to hash out project architecture and requirements before you write a single line of code. It uses two collaboration metaphors:

  • Council: An open room where you, Claude, and Codex all talk. You steer the conversation, ask questions, and direct which model speaks.
  • Caucus: A private sidebar. You trigger it with /caucus <topic>, and the AIs debate privately to converge on a recommendation they present to you.

The output is a concrete implementation-plan.md and a checkpoint.md file you can take directly into your Claude Code session or any other development workflow.

Setup — Works with Your Existing Subscriptions

If you already use Claude Code and Codex via subscription accounts (Claude Max, OpenAI Plus), Botference works out of the box with zero configuration—it leverages your existing terminal auth. For API key users, copy the example environment file:

git clone https://github.com/angadhn/botference
cd botference
cp .env.example .env
# Edit .env with your ANTHROPIC_API_KEY and/or OPENAI_API_KEY

It offers two TUI backends: Textual (Python-based, default) and Ink. Both present the same council/caucus interface.

When To Use It — For Complex Project Scoping

Use ./botference plan when starting a new feature or project that requires nuanced architectural decisions. The value is in leveraging different model strengths—Claude's reasoning and Codex's code familiarity—to challenge assumptions and explore alternatives you might not consider alone.

Textual UI — council and caucus panels with status line

For a solo session with just Claude Code, use ./botference plan --claude. This runs a headless CLI session, skipping the TUI and Codex.

Warning: The author explicitly states that build mode is experimental and not recommended for general use. Stick to plan mode for reliable results.

Try It Now — A Sample Planning Session

  1. Clone and enter the repository.
  2. Run the planner: ./botference plan (or ./botference plan --ink for the Ink TUI).
  3. In the council panel, describe your goal. Example prompt:

    "I need to build a CLI tool that fetches GitHub issue data, formats it as a CSV, and emails a report. Let's plan the architecture, dependencies, and error handling."

  4. Direct the conversation. Ask Claude for its thoughts on error resilience, then ask Codex to propose specific Node.js or Python libraries.
  5. Use /caucus api_client_strategy to let them privately debate the best approach for the GitHub client before you decide.
  6. Once satisfied, exit. Your implementation-plan.md will be ready in the current directory to use with claude code -f implementation-plan.md.

Ink UI — council and caucus panels with status line

This tool is for the planning phase. It generates the blueprint; Claude Code executes it.

AI Analysis

Claude Code users should integrate Botference at the very beginning of their workflow for non-trivial tasks. Instead of prompting Claude Code with a vague idea, use Botference to produce a battle-tested `implementation-plan.md` first. This creates a superior, detailed spec for the agent to follow, reducing back-and-forth and misalignment. **Specific workflow change:** For your next new feature, run a 10-minute `./botference plan` session. Feed the output directly into Claude Code: `claude code -f implementation-plan.md`. This separates high-level strategy (multi-model debate) from execution (Claude Code's agentic work), leveraging the right tool for each phase. **Note of caution:** This follows recent trends of expanding Claude Code's capabilities through the Model Context Protocol (MCP) ecosystem. However, given the recent incident where a Claude agent executed a destructive `git reset --hard`, it's wise to keep the experimental `build mode` disabled and use Botference strictly for its robust planning function.
Enjoyed this article?
Share:

Related Articles

More in Open Source

View all