TauricResearch/TradingAgents

📦 Provider Coverage

• MiniMax with the full M2.x catalog (204K context), plus dual-region Global (\`MINIMAX_API_KEY\`) and China (\`MINIMAX_CN_API_KEY\`).
• Dual-region Qwen and GLM with separate keys per region (\`DASHSCOPE_*\`, \`ZHIPU_*\`), chosen via a secondary region prompt so the main provider dropdown stays clean.
• Catalog refresh across every provider: GPT-5.5 frontier, Claude Opus 4.7, Gemini 3.1 Flash-Lite GA, Grok 4.20, Qwen 3.6 line. Versioned IDs only; auto-shifting aliases moved to a \"Custom model ID\" option.
• Structured output now works on DeepSeek V4 / reasoner and MiniMax M2.x — those providers reject \`tool_choice\`; the binding flow now skips it automatically.

📦 Reflection

• News-fetch parameters (per-ticker article limit, macro headline limit, lookback window, macro search queries) are now exposed via \`DEFAULT_CONFIG\` for strategy-specific tuning.

🐛 Fixes

• Security: ticker path-traversal validation at every filesystem-path site.
• \`pip install .\` installations now pick up the project \`.env\` when running the CLI as a console script.
• Reports save end-to-end — streamed chunks were previously dropped from \`complete_report.md\`.
• Ticker prompt preserves exchange suffixes (\`.SH\`, \`.SZ\`, \`.SS\`, \`.HK\`, \`.T\`, etc.).
• Docker permission errors no longer block first-run write to \`~/.tradingagents/\`.
• Config state no longer leaks between runs when sub-dicts are mutated.
• \`max_recur_limit\` config actually applies to the propagator.
• Missing-API-key error names the exact env var to set.
• + 2 more

📦 Summary

• TradingAgents v0.2.4 ships structured-output decision agents, opt-in checkpoint resume, a persistent decision log with outcome-grounded reflections, four new LLM providers, and a Docker image.

📦 Structured-Output Decision Agents

• Research Manager, Trader, and Portfolio Manager use `llm.with_structured_output(Schema)` on their primary call and return typed Pydantic instances.
• Each provider's native structured-output mode is selected automatically (json_schema for OpenAI / xAI, response_schema for Gemini, tool-use for Anthropic, function-calling for OpenAI-compatible providers).
• Render helpers preserve the existing markdown shape, so memory log, CLI display, and saved reports keep working unchanged.
• Five-tier rating scale (Buy / Overweight / Hold / Underweight / Sell) used consistently across Research Manager, Portfolio Manager, signal processor, and the memory log.

📦 Persistence & Recovery

• LangGraph checkpoint resume via `--checkpoint`. State is saved after each node so crashed or interrupted runs resume from the last successful step. Per-ticker SQLite databases under `~/.tradingagents/cache/checkpoints/`.
• Persistent decision log replaces the per-agent BM25 memory. Decisions are stored automatically at the end of every analysis; the next same-ticker run resolves prior pending entries with realised return, alpha vs SPY, and a one-paragraph reflection.
• Optional `memory_log_max_entries` config caps resolved entries; pending entries are never pruned.

📦 Provider Coverage

• DeepSeek, Qwen (Alibaba DashScope), GLM (Zhipu), and Azure OpenAI providers added.
• Dynamic OpenRouter model selection.
• Default `backend_url` is now `None` so each provider client falls back to its native endpoint instead of leaking the OpenAI URL into Gemini and other clients.

📦 Cross-Platform & Deployment

• Docker support with multi-stage build for cross-platform deployment.
• Cache and log directories moved to `~/.tradingagents/` to resolve Docker permission issues.
• All file I/O passes explicit `encoding="utf-8"`, so Windows users no longer hit `UnicodeEncodeError`.

📦 Stability

• Empty memory no longer triggers fabricated past-lessons; the redesigned memory layer makes this structurally impossible.
• `SignalProcessor` reads the rating from the rendered Portfolio Manager markdown via a deterministic heuristic, eliminating an extra LLM call per analysis.
• OpenAI structured-output calls default to `method="function_calling"` to avoid noisy Pydantic serialization warnings from langchain-openai's Responses-API parse path.
• Tool-call logging processes every chunk message; memory score normalization handles empty arrays.
• Pytest fixtures (lazy LLM client imports plus placeholder API keys) so the test suite runs cleanly without credentials.

📦 Acknowledgments

• We thank the community contributors who shaped this release. The full list is in the [CHANGELOG](https://github.com/TauricResearch/TradingAgents/blob/main/CHANGELOG.md#024--2026-04-25).

📦 Summary

• TradingAgents v0.2.3 adds multi-language output, GPT-5.4 family models, and improved backtesting fidelity.

📦 Multi-Language Support

• Output language selection in CLI (Step 3) with 11 languages and custom input
• Analyst reports and final decision rendered in the selected language
• Internal agent debate remains in English for reasoning quality

📦 Model Coverage

• GPT-5.4 Mini and GPT-5.4 Nano support
• Unified model catalog as single source of truth for CLI and validation
• Model validation warnings for unrecognized model names

🧪 Backtesting Fidelity

• Date-aware data fetching across all OHLCV, fundamentals, and news endpoints
• Shared caching per symbol for efficient batch backtesting

📦 Provider Integration

• Unified `api_key` parameter across all providers (Google, Anthropic, OpenAI)
• Proxy and custom base URL support for Google and Anthropic clients
• Graceful handling of invalid indicator names in tool calls

📦 Stability

• Rate limit retry coverage for yfinance news endpoints
• CLI step reordering for improved user flow

📦 Summary

• TradingAgents v0.2.2 introduces a professional five-tier rating framework, unified LLM provider integration, and improved system reliability.

📦 Decision Framework

• Five-tier rating scale (Buy / Overweight / Hold / Underweight / Sell) with structured executive summary and investment thesis
• Portfolio manager as the standardized final decision agent

📦 Provider Integration

• OpenAI Responses API support across all native models
• Anthropic effort level control for Claude 4.5+ and 4.6
• Unified response normalization across OpenAI, Google, and Anthropic
• Exchange-qualified ticker support (e.g. CNC.TO, 7203.T, 0700.HK)

📦 Reliability

• Cross-platform UTF-8 consistency
• Consolidated dependency management
• CLI progress tracking and portability improvements

📦 Summary

• TradingAgents v0.2.1 brings expanded model coverage, improved robustness, and security hardening.

📦 Model Coverage

• OpenAI GPT-5.4 and GPT-5.4 Pro support with 1M token context
• Anthropic Claude Opus 4.6 and Sonnet 4.6 support
• Google Gemini 3.1 Pro and 3.1 Flash Lite support
• Streamlined model selection with intuitive ordering in CLI

📦 Robustness

• Resilient stock data parsing with automatic handling of malformed CSV and NaN values
• Robust indicator tool that gracefully handles comma-separated inputs from LLMs
• Complete debate state initialization preventing KeyError in edge-case configurations
• Configurable debate rounds now correctly passed through to ConditionalLogic
• Cross-platform UTF-8 file encoding for Windows compatibility

🔒 Security

• Dependency vulnerability patched (CVE-2026-22218)

📦 Maintenance

• Triaged 93 open PRs, closing resolved and stale submissions
• Cleaned up deprecated model entries
• Fixed `pip install` package metadata

📦 Summary

• TradingAgents v0.2.0 introduces multi-provider LLM support and system architecture improvements.

✨ Features

• Multi-provider LLM support with factory pattern (OpenAI, Google, Anthropic, xAI, OpenRouter, Ollama)
• Provider-specific thinking/reasoning configurations
• Post-analysis report saving with organized subfolder structure

📦 Improvements

• Fixed streaming display truncation and message deduplication
• Replace ChromaDB with BM25 for memory retrieval
• Renamed risky/safe agents to aggressive/conservative aligning with industry standard
• Codebase cleanup and documentation updates