◈──◆──◇ GIIZHENDAM AABAJICHIGANAN MCP SERVER ◇──◆──◈

"Giizhendam Aabajichiganan" (Ojibwe: Decision-Making Tools) - An MCP server providing AI-assisted programming and decision-making tools.

ᐴ WAAWIINDAMAAGEWIN ᔔ [OVERVIEW] ◈──◆──◇──◆──◈

This project implements a Model Context Protocol (MCP) server that provides multi-agent CLI orchestration for developers and decision-makers. It serves as a bridge between different AI models and external CLI agents, and provides specialized tools for financial analysis and collaborative decision simulation.

The server includes a multi-agent orchestrator tool (orchestrate_agents) that loads agent definitions from CLI_AGENTS_JSON or llms.txt. It uses a configurable orchestrator model (default: deepseek/deepseek-v3.2-speciale) via Zenmux or any OpenAI-compatible provider to decide between sequential or parallel execution, executes the agents using the Qwen CLI (running off z-ai/glm-4.6v-flash), and synthesizes a consolidated markdown report.

Key Features:

• Flexible Provider Support: Seamlessly switch between Zenmux (default) or any OpenAI-compatible provider (e.g., OpenRouter, OpenAI, etc.)
• Unified Orchestrator Model: All AI operations use a single configurable model
• Qwen CLI Integration: Leverages qwen -y with configurable model (default: z-ai/glm-4.6v-flash)
• 18 Financial Expert Agents: Comprehensive financial analysis with individual expert perspectives (unlimited tokens) + RAG consolidation
• Model Flexibility: Easily swap between any Zenmux-supported model

ᐴ GASHKITOONAN ᔔ [CAPABILITIES] ◈──◆──◇──◆──◈

• ◇ Multi-Agent Orchestrator ◇

  • Load agent personas/instructions from CLI_AGENTS_JSON or llms.txt
  • Model-chosen sequential or parallel execution
  • Uses qwen -y for autonomous task execution
  • Consolidated markdown synthesis of agent outputs

• ◇ Financial Expert Simulation ◇

  • Orchestrates 18 financial expert agents individually (Comprehensive analysis - no token limits)
  • Uses orchestrator model with File Search RAG to consolidate all expert outputs (16k token context)
  • Generates enterprise-ready, production-grade analysis and strategic advisory
  • Includes perspectives from: Damodaran, Graham, Buffett, Munger, Lynch, Fisher, Ackman, Wood, Burry, Pabrai, Jhunjhunwala, Druckenmiller, plus Valuation, Sentiment, Fundamentals, Technicals, Risk Manager, and Portfolio Manager agents

• ◇ CEO & Board Meeting Simulation ◇

  • Simulates realistic board discussions on specified topics
  • Customizable roles and discussion parameters
  • Useful for planning and decision-making exercises

• ◇ Security & Reliability ◇

  • Built-in security checks for file operations and URL validation
  • Comprehensive error handling and logging
  • Secure API key management

ᐴ OSHKI-AABAJICHIGANAN ᔔ [RECENT CHANGES] ◈──◆──◇──◆──◈

• v0.6.8 - Provider Flexibility: Added support for generic OpenAI-compatible providers (OpenAI, OpenRouter, etc.) via OPENAI_BASE_URL and OPENAI_API_KEY, while maintaining Zenmux as the default.
• v0.6.7 - Enterprise Enhancement: Upgraded finance_experts to use all 18 experts with comprehensive analysis (no token limits) and 16k context consolidation.
• v0.6.6 - Major Refactor: Switched orchestrator provider to Zenmux (OpenAI SDK compatible). Replaced Aider/generic CLI with Qwen CLI (qwen -y) for agent execution. CLI_AGENTS_JSON now defines agent personas/instructions rather than raw commands.
• v0.6.0 - Unified orchestrator model configuration.
• v0.5.3 - Added 18 financial expert agents with dynamic prompt loading from markdown files
• v0.3.34 - Fixed shebang line in the bundled output file to ensure proper execution via npx.

ᐴ NITAM-AABAJICHIGANAN ᔔ [PREREQUISITES] ◈──◆──◇──◆──◈

• Node.js (v14 or higher) and npm/yarn
• Qwen CLI (qwen) installed and available in PATH (configured for yolo mode support)
• Zenmux API Key or OpenAI API Key (required) - Used for all AI operations
• ORCHESTRATOR_MODEL (optional) - Defaults to deepseek/deepseek-v3.2-speciale

ᐴ AABAJITOOWINAN ᔔ [INSTALLATION] ◈──◆──◇──◆──◈

Global Installation (Recommended)

╭──────────────────────────────────────────────────────────────────────╮
│  ᐴ AABAJITOOWINAN ᔔ [ IMPLEMENTATION COMMANDS ]                      │
╰──────────────────────────────────────────────────────────────────────╯

# First uninstall any existing versions
npm uninstall -g @nbiish/giizhendam-aabajichiganan-mcp
npm uninstall -g @nbiish/ai-tool-mcp

# Install from the current directory
npm install -g .

# Or install from npm registry via npx (recommended for MCP)
# This is the preferred method for MCP configuration
npx -y @nbiish/giizhendam-aabajichiganan-mcp

# Or install globally from npm registry
npm install -g @nbiish/giizhendam-aabajichiganan-mcp
# Alternative package with identical functionality
npm install -g @nbiish/ai-tool-mcp

Troubleshooting Installation Issues

If you encounter issues with the executable script (e.g., "command not found" or shell syntax errors), try these steps:

╭──────────────────────────────────────────────────────────────────────╮
│  ᐴ NANAA'ITOOWIN ᔔ [ TROUBLESHOOTING COMMANDS ]                      │
╰──────────────────────────────────────────────────────────────────────╯

# Check if the installed script has the proper shebang line:
cat $(which giizhendam-mcp)

# If the shebang line is missing or incorrect, fix it manually:
echo '#!/usr/bin/env node' > /tmp/fixed-script
cat $(which giizhendam-mcp) >> /tmp/fixed-script
sudo mv /tmp/fixed-script $(which giizhendam-mcp)
sudo chmod +x $(which giizhendam-mcp)

# Alternatively, reinstall after clearing npm cache:
npm cache clean --force
npm uninstall -g @nbiish/giizhendam-aabajichiganan-mcp
npm install -g .

ᐴ ONAAKONIGE ᔔ [CONFIGURATION] ◈──◆──◇──◆──◈

MCP Client Configuration (mcp.json)

Configure the server in your MCP client's configuration file. The location depends on your client:

• Cursor: ~/.cursor/mcp.json or ~/.config/cursor/mcp.json
• Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows)
• Other MCP Clients: Check your client's documentation

Basic Configuration (Recommended)

{
  "mcpServers": {
    "giizhendam-mcp": {
      "command": "npx",
      "args": ["-y", "@nbiish/giizhendam-aabajichiganan-mcp"],
      "env": {
        "ZENMUX_API_KEY": "sk-your-zenmux-key-here",
        "ORCHESTRATOR_MODEL": "deepseek/deepseek-v3.2-speciale"
      }
    }
  }
}

Configuration with OpenAI Provider (OpenRouter/Other)

{
  "mcpServers": {
    "giizhendam-mcp": {
      "command": "npx",
      "args": ["-y", "@nbiish/giizhendam-aabajichiganan-mcp"],
      "env": {
        "OPENAI_BASE_URL": "https://openrouter.ai/api/v1",
        "OPENAI_API_KEY": "sk-your-openrouter-key-here",
        "ORCHESTRATOR_MODEL": "anthropic/claude-3-opus"
      }
    }
  }
}

Full Configuration with All Options

{
  "mcpServers": {
    "giizhendam-mcp": {
      "command": "npx",
      "args": ["-y", "@nbiish/giizhendam-aabajichiganan-mcp"],
      "env": {
        "ZENMUX_API_KEY": "sk-your-zenmux-key-here",
        "ORCHESTRATOR_MODEL": "deepseek/deepseek-v3.2-speciale",
        "CLI_AGENTS_JSON": "[{\"name\":\"FinanceExpert\",\"cmd\":\"You are an expert financial analyst.\"}]",
        "AGENT_OUTPUT_DIR": "./output/agents",
        "AGENT_MODEL": "z-ai/glm-4.6v-flash",
        "EXECUTION_STYLE": "auto",
        "FINANCE_EXPERTS_OUTPUT_DIR": "./output/finance-experts",
        "CEO_BOARD_OUTPUT_DIR": "./output/ceo-and-board",
        "SYNTH_MAX_PER_AGENT_CHARS": "20000",
        "SYNTH_MAX_TOTAL_CHARS": "150000"
      }
    }
  }
}

Environment Variables Reference

Required Variables

Variable	Description	Example
ZENMUX_API_KEY	Your Zenmux API key (required if not using generic OpenAI provider)	sk-...
OPENAI_API_KEY	Generic OpenAI/OpenRouter API key (alternative to Zenmux)	sk-...
OPENAI_BASE_URL	Base URL for generic OpenAI provider (e.g. OpenRouter)	https://openrouter.ai/api/v1

Optional Variables

Variable	Default	Description
ORCHESTRATOR_MODEL	deepseek/deepseek-v3.2-speciale	Model used for orchestrator.
AGENT_MODEL	z-ai/glm-4.6v-flash	Model used for Qwen CLI agents.
CLI_AGENTS_JSON	(none)	JSON array of agent definitions (defines persona/instructions for Qwen CLI)
AGENT_OUTPUT_DIR	./output/agents	Directory for orchestrator tool outputs
EXECUTION_STYLE	auto	Execution style: auto (model decides), sequential, or parallel
FINANCE_EXPERTS_OUTPUT_DIR	./output/finance-experts	Directory for financial expert analysis outputs
CEO_BOARD_OUTPUT_DIR	./output/ceo-and-board	Directory for board simulation outputs
SYNTH_MAX_PER_AGENT_CHARS	20000	Max characters per agent in synthesis
SYNTH_MAX_TOTAL_CHARS	150000	Max total characters in synthesis

Orchestrator Model Options

The ORCHESTRATOR_MODEL can be any model supported by Zenmux. Popular options:

DeepSeek Models:

• deepseek/deepseek-v3.2-speciale (default)
• deepseek/deepseek-coder-v2

OpenAI Models:

• gpt-4o
• gpt-4-turbo

Z-AI Models (for Agents):

• z-ai/glm-4.6v-flash (default for agents)

CLI Agents Configuration

Configure agent personas for the orchestrate_agents tool in two ways:

Option A: Using CLI_AGENTS_JSON (Recommended)

Set the CLI_AGENTS_JSON environment variable as a JSON array. Each entry defines an agent "persona" that will be executed via qwen -y. The cmd field now represents the System Instruction or Persona for that agent.

[
  {"name": "SeniorDev", "cmd": "You are a senior developer expert in TypeScript and Node.js."},
  {"name": "SecurityAudit", "cmd": "You are a security auditor looking for vulnerabilities."},
  {"name": "RefactorSpec", "cmd": "You specialize in code refactoring and clean architecture."}
]

Important: Escape quotes properly in JSON.

Option B: Using llms.txt (Fallback)

If CLI_AGENTS_JSON is not set, the server reads llms.txt from the current working directory. The command block should contain the persona instructions.

- SeniorDev
```bash
You are a senior developer expert in TypeScript and Node.js.

• SecurityAudit

You are a security auditor looking for vulnerabilities.

### Output Directories


All output directories are relative to the server's current working directory (typically your project root). The server will create these directories automatically if they don't exist.

**Default Structure:**

your-project/ ├── output/ │ ├── agents/ # Orchestrator tool outputs │ ├── finance-experts/ # Financial expert analyses │ │ └── expert_outputs_*/ # Individual expert files │ └── ceo-and-board/ # Board simulation outputs

### Configuration Examples

#### Minimal Configuration (Finance Experts Only)

```json
{
  "mcpServers": {
    "giizhendam-mcp": {
      "command": "npx",
      "args": ["-y", "@nbiish/giizhendam-aabajichiganan-mcp"],
      "env": {
        "ZENMUX_API_KEY": "your-key-here"
      }
    }
  }
}

With Custom Model

{
  "mcpServers": {
    "giizhendam-mcp": {
      "command": "npx",
      "args": ["-y", "@nbiish/giizhendam-aabajichiganan-mcp"],
      "env": {
        "ZENMUX_API_KEY": "your-key-here",
        "ORCHESTRATOR_MODEL": "anthropic/claude-3.5-sonnet"
      }
    }
  }
}

With CLI Agents

{
  "mcpServers": {
    "giizhendam-mcp": {
      "command": "npx",
      "args": ["-y", "@nbiish/giizhendam-aabajichiganan-mcp"],
      "env": {
        "ZENMUX_API_KEY": "your-key-here",
        "CLI_AGENTS_JSON": "[{\"name\":\"CodeArchitect\",\"cmd\":\"You are a software architect.\"}]"
      }
    }
  }
}

Important Notes

⚠️ Security:

• Never commit your mcp.json file with real API keys to version control
• Use environment variables or secret management tools in production

⚠️ Breaking Changes (v0.6.6):

• Switched from OpenRouter to Zenmux/OpenAI
• Replaced Aider/generic CLI support with dedicated Qwen CLI support (qwen -y)
• CLI_AGENTS_JSON semantics changed: cmd is now a system instruction, not a shell command template

Troubleshooting Configuration

Issue: "ZENMUX_API_KEY is not set"

• Solution: Ensure ZENMUX_API_KEY is set in the env section of your mcp.json (or OPENAI_API_KEY)

Issue: "Model not found"

• Solution: Verify the model name matches Zenmux/OpenAI format (e.g., gpt-4o)

Issue: "No CLI agents configured"

• Solution: Set CLI_AGENTS_JSON or create llms.txt in your project directory

Issue: Output directories not created

• Solution: Ensure the server has write permissions in the current working directory

ᐴ INAABAJICHIGAN ᔔ [USAGE] ◈──◆──◇──◆──◈

Orchestrate CLI Agents

Use the orchestrate_agents tool to run your configured CLI agents and produce a consolidated report:

╭──────────────────────────────────────────────────────────────────────╮
│  ᐴ WIIDOOKAAZOWIN ᔔ [ CODE ASSISTANCE ]                           │
╰──────────────────────────────────────────────────────────────────────╯

const result = await server.execute("orchestrate_agents", {
  prompt_text: "Create a React component that displays a counter with increment and decrement buttons"
});

// Returns a summary with execution style, per-agent outputs, and synthesis markdown path in _meta

Notes

The former double computation tool has been removed. If you need redundant verification, you can invoke orchestrate_agents multiple times and compare outputs.

Financial Expert Simulation

Orchestrates 18 financial expert agents individually, then uses the orchestrator model with File Search RAG to consolidate all perspectives into enterprise-ready analysis:

╭──────────────────────────────────────────────────────────────────────╮
│  ᐴ ZHOONIYAAWICHIGEWIN ᔔ [ FINANCIAL EXPERTISE ]                     │
╰──────────────────────────────────────────────────────────────────────╯

const result = await server.execute("finance_experts", {
  topic: "Funding strategy for our new AI product line",
  output_filename: "ai_product_funding_analysis"  // Optional: custom filename
});

// Process:
// 1. Each of 18 experts provides analysis (900 tokens each)
// 2. Individual expert files saved to expert_outputs_[timestamp]/
// 3. Orchestrator model consolidates all outputs using File Search RAG
// 4. Results saved to ./finance-experts/ai_product_funding_analysis_[timestamp].md
//    - Includes all individual expert perspectives
//    - Includes consolidated RAG-based analysis
//    - Includes recommended orchestrator prompt for CLI tools

Board Meeting Simulation

Simulate a board discussion on a specific topic:

╭──────────────────────────────────────────────────────────────────────╮
│  ᐴ MAAWANJIDIWIN ᔔ [ BOARD MEETING ]                                 │
╰──────────────────────────────────────────────────────────────────────╯

const result = await server.execute("ceo_and_board", {
  topic: "Q3 Strategy Review: Expansion into European Markets",
  output_filename: "q3_europe_expansion_board_meeting"  // Optional: custom filename
});

// The tool now uses standard board roles by default, including:
// Board Chair, CEO, CFO, COO, CTO, Independent Director, Corporate Secretary/General Counsel, etc.

// You can also specify custom roles if needed:
const result = await server.execute("ceo_and_board", {
  topic: "Q3 Strategy Review: Expansion into European Markets",
  roles: ["CEO", "CFO", "CTO", "Lead Investor", "Independent Director"],
  output_filename: "q3_europe_expansion_board_meeting"
});

// Results saved to the configured output directory or default ./ceo-and-board/

ᐴ WAABANDA'IWEWIN ᔔ [EXAMPLES] ◈──◆──◇──◆──◈

Anishinaabe Cultural Examples

Here are some examples of how these tools could be applied in contexts relevant to Anishinaabe culture and communities:

╭──────────────────────────────────────────────────────────────────────╮
│  ᐴ ANISHINAABE INAADIZIWIN ᔔ [ CULTURAL APPLICATIONS ]               │
╰──────────────────────────────────────────────────────────────────────╯

// Example using orchestrator for language revitalization
const result = await server.execute("orchestrate_agents", {
  prompt_text: "Help draft an Ojibwe language localization file (oj.json) for our UI based on this English template file. Ensure respectful and accurate translations.",
});

// Example: run orchestrator for resource management simulation
const result2 = await server.execute("orchestrate_agents", {
  prompt_text: "Verify the logic in wild_rice_harvest_simulation.js for sustainable yield calculations reflecting traditional ecological knowledge principles."
});

// Example using finance_experts for community project planning
const result = await server.execute("finance_experts", {
  topic: "Financial sustainability plan for the Ojibwe Language Nest startup, considering long-term community benefit and grant opportunities.",
  output_filename: "ojibwe_language_nest_finance_plan"
});

// Example using ceo_and_board for governance simulation
const result = await server.execute("ceo_and_board", {
  topic: "Tribal Council discussion on investing treaty settlement funds into renewable energy infrastructure on reservation lands, balancing economic development with environmental stewardship.",
  roles: ["Tribal Chair", "Treasurer", "Council Member (Economic Development)", "Council Member (Natural Resources)", "Elder Advisor", "Community Member Rep"],
  output_filename: "treaty_fund_renewable_energy_discussion"
});

Tool Reference

orchestrate_agents

Parameter	Type	Description	Required
prompt_text	string	The main prompt/instruction that is passed to each CLI agent’s command template	Yes

finance_experts

Parameter	Type	Description	Required
topic	string	The central financial topic or query related to a project or business situation	Yes
output_filename	string	Optional filename (without extension) for the output markdown file	No

ceo_and_board

Parameter	Type	Description	Required
topic	string	The central topic for the board discussion	Yes
roles	string[]	Optional list of board member roles to simulate. If not provided, standard board roles will be used	No
output_filename	string	Optional filename (without extension) for the output markdown file	No

Project Structure

giizhendam-aabajichiganan-mcp/
├── src/                    # Source code
│   └── index.ts            # Main server implementation
├── dist/                   # Compiled JavaScript output
├── output/                 # Configurable output directory (example)
│   ├── agents/             # Per-agent outputs and synthesis markdown
│   ├── finance-experts/    # Financial expert simulation output
│   └── ceo-and-board/      # Board simulation output
├── package.json            # Project metadata and dependencies
└── tsconfig.json           # TypeScript configuration

Development

# Run in development mode
npm run dev

# Build
npm run build

# Test
npm test

Local Testing Plan

• Prepare env
  • Set ZENMUX_API_KEY in your MCP config env or shell (required).
  • Optionally set ORCHESTRATOR_MODEL to test different models (defaults to gpt-4o).
  • Optionally set CLI_AGENTS_JSON or populate llms.txt for orchestrator tool.
  • Ensure qwen CLI is installed and in PATH.
• Install and Test
  • Test via npx: npx -y @nbiish/giizhendam-aabajichiganan-mcp
  • Or install globally: npm install -g @nbiish/giizhendam-aabajichiganan-mcp
  • Launch via your MCP client using the mcp.json example above.
• Invoke tools
  • Orchestrator: Call orchestrate_agents with a short prompt_text.
    • Expected: Server selects parallel or sequential, executes Qwen agent, writes per-agent files under AGENT_OUTPUT_DIR, and creates a synthesis markdown file.
  • Finance Experts: Call finance_experts with a financial topic.
    • Expected: 18 expert analyses (900 tokens each) saved individually, then consolidated via RAG into enterprise-ready analysis.
  • Board Simulation: Call ceo_and_board with a discussion topic.
    • Expected: Realistic board meeting transcript and recommended orchestrator prompt.
• Verify outputs
  • Confirm files exist in configured output directories and review reports.
  • Adjust EXECUTION_STYLE to sequential or parallel to force behavior.
  • Test different ORCHESTRATOR_MODEL values to compare outputs.
• Troubleshoot
  • If timeouts occur, tune SYNTH_MAX_PER_AGENT_CHARS, SYNTH_MAX_TOTAL_CHARS.
  • Ensure qwen command runs successfully from your shell with qwen -y "test".
  • Check Zenmux/OpenAI dashboard for API usage.

Citation

@misc{giizhendam-aabajichiganan-mcp2025,
  author/creator/steward = {ᓂᐲᔥ ᐙᐸᓂᒥᑮ-ᑭᓇᐙᐸᑭᓯ (Nbiish Waabanimikii-Kinawaabakizi), also known legally as JUSTIN PAUL KENWABIKISE, professionally documented as Nbiish-Justin Paul Kenwabikise, Anishinaabek Dodem (Anishinaabe Clan): Animikii (Thunder), descendant of Chief ᑭᓇᐙᐸᑭᓯ (Kinwaabakizi) of the Beaver Island Band and enrolled member of the sovereign Grand Traverse Band of Ottawa and Chippewa Indians},
  title/description = {giizhendam-aabajichiganan-mcp},
  type_of_work = {Indigenous digital creation/software incorporating traditional knowledge and cultural expressions},
  year = {2025},
  publisher/source/event = {GitHub repository under tribal sovereignty protections},
  howpublished = {\url{https://github.com/nbiish/giizhendam-aabajichiganan-mcp}},
  note = {Authored and stewarded by ᓂᐲᔥ ᐙᐸᓂᒥᑮ-ᑭᓇᐙᐸᑭᓯ (Nbiish Waabanimikii-Kinawaabakizi), also known legally as JUSTIN PAUL KENWABIKISE, professionally documented as Nbiish-Justin Paul Kenwabikise, Anishinaabek Dodem (Anishinaabe Clan): Animikii (Thunder), descendant of Chief ᑭᓇᐙᐸᑭᓯ (Kinwaabakizi) of the Beaver Island Band and enrolled member of the sovereign Grand Traverse Band of Ottawa and Chippewa Indians. This work embodies Indigenous intellectual property, traditional knowledge systems (TK), traditional cultural expressions (TCEs), and associated data protected under tribal law, federal Indian law, treaty rights, Indigenous Data Sovereignty principles, and international indigenous rights frameworks including UNDRIP. All usage, benefit-sharing, and data governance are governed by the COMPREHENSIVE RESTRICTED USE LICENSE FOR INDIGENOUS CREATIONS WITH TRIBAL SOVEREIGNTY, DATA SOVEREIGNTY, AND WEALTH RECLAMATION PROTECTIONS.}
}

License

This project is licensed under the terms specified in the LICENSE file. This license is a COMPREHENSIVE RESTRICTED USE LICENSE FOR INDIGENOUS CREATIONS WITH TRIBAL SOVEREIGNTY, DATA SOVEREIGNTY, AND WEALTH RECLAMATION PROTECTIONS.

Contributing

Please read CONTRIBUTING.md for details on our code of conduct and the process for submitting pull requests.

Terms of Service

The authoritative Terms of Service for this codebase MUST always be synchronized with:

• https://raw.githubusercontent.com/nbiish/license-for-all-works/refs/heads/main/Terms-of-Service.md

Privacy Policy

The authoritative Privacy Policy for this codebase MUST always be synchronized with:

• https://raw.githubusercontent.com/nbiish/license-for-all-works/refs/heads/main/Privacy-Policy.md

Copyright © 2025 ᓂᐲᔥ ᐙᐸᓂᒥᑮ-ᑭᓇᐙᐸᑭᓯ (Nbiish Waabanimikii-Kinawaabakizi), also known legally as JUSTIN PAUL KENWABIKISE, professionally documented as Nbiish-Justin Paul Kenwabikise, Anishinaabek Dodem (Anishinaabe Clan): Animikii (Thunder), a descendant of Chief ᑭᓇᐙᐸᑭᓯ (Kinwaabakizi) of the Beaver Island Band, and an enrolled member of the sovereign Grand Traverse Band of Ottawa and Chippewa Indians. This work embodies Traditional Knowledge and Traditional Cultural Expressions. All rights reserved.

Release Notes

v0.6.6 (Current)

• Major Refactor: Unified orchestrator model configuration via Zenmux (OpenAI SDK)
• Replaced Aider/generic CLI support with dedicated Qwen CLI support (qwen -y)
• Default Models Updated:
  • Orchestrator: deepseek/deepseek-v3.2-speciale
  • Agents: z-ai/glm-4.6v-flash (via qwen -m)
• Enhanced CLI_AGENTS_JSON to define personas/instructions for Qwen
• Supports any Zenmux-supported model (OpenAI, Anthropic, Google)
• Updated MCP configuration for simpler usage

v0.6.0

• Major Refactor: Unified orchestrator model configuration via OpenRouter
• Removed direct Gemini API dependency (@google/generative-ai)
• All AI operations now use single configurable ORCHESTRATOR_MODEL (default: google/gemini-2.5-pro)
• Enhanced finance_experts tool: 18 expert agents, 900-token limits, RAG consolidation
• Supports any OpenRouter-supported model (Gemini, Claude, GPT-4, etc.)
• Updated MCP configuration to use npx -y format for easier installation
• Added comprehensive orchestrator model configuration documentation

v0.5.3

• Added 18 financial expert agents with dynamic prompt loading from markdown files
• Enhanced expert prompts with detailed research and methodologies

v0.3.34

• Fixed shebang line in the bundled output file to ensure proper execution via npx
• Resolved "Client closed" errors when running via MCP