TradingAgents

Changelog

All notable changes to TradingAgents are documented here.

The format is based on Keep a Changelog, and this project follows Semantic Versioning. Breaking changes within the 0.x line are called out explicitly.

[0.3.1] — 2026-07-05

Correctness and stability patch: data look-ahead, graph-router crash-safety, checkpoint identity, crypto sentiment sources, and configurable resilience.

Fixed

• Alpha Vantage look-ahead filter now runs. The fundamentals payload is a JSON string, so the dict-only guard skipped filtering and future-dated reports leaked into historical runs; parse before filtering. (#1115, @zachthebird)
• News analyst prompt matches the tool. The prompt advertised get_news(query, ...) but the tool takes a ticker; aligned to stop hallucinated free-text query calls. (#1116, @shcheuk)
• Shared debate/risk routers can’t crash mid-run. Both routers return more targets than any one edge mapped; every edge now shares the complete path map, so a fall-through under prompt/i18n/refactor drift stays routable. (#1088, @Fr3ya, @sa7an7, @Sushanth012)
• Checkpoint resume respects graph shape. The thread id folds in selected analysts, debate/risk depth, and asset mode, so a resume under different choices no longer continues the wrong graph. (#1089, @bossjoker1, @Ghraven)
• Crypto sentiment sources resolve. StockTwits lists crypto as <BASE>.X (Yahoo’s BTC-USD 404s) and Reddit needs the base symbol to match; the social path now maps crypto correctly for both. (#1113, @suremadoreai)

Added

• Configurable LLM retry budget. llm_max_retries / TRADINGAGENTS_LLM_MAX_RETRIES is forwarded to every provider, so a transient 429 burst no longer aborts a run. (#1091, @yanggaome)
• Bedrock API-key auth. AWS_BEARER_TOKEN_BEDROCK authenticates Amazon Bedrock without AWS access keys and takes precedence over an ambient AWS_PROFILE. (#1103, @praxstack)
• Latest Claude models. Added Claude Sonnet 5 (claude-sonnet-5) and Fable 5 (claude-fable-5); effort control now covers the Claude 5 line.

[0.3.0] — 2026-06-22

Stabilization and extensibility release: a CI gate, a unified verified data-access contract, a provider and data-vendor registry, and a maintenance sweep that hardened config precedence, the model catalog, data resilience, and structured output.

Added

• CI gate. GitHub Actions runs the pytest suite across Python 3.10-3.13, strict ruff, and a clean-install smoke that imports the package and CLI to catch undeclared dependencies. (#994, #197)
• Provider registry. OpenAI-compatible providers register as a single spec, and a generic openai_compatible endpoint covers vLLM, LM Studio, and relays. Adds NVIDIA NIM, Kimi, Groq, Mistral, and a native Amazon Bedrock client.
• Macro and prediction-market vendors. FRED macro indicators and Polymarket event probabilities, surfaced to the news and macro analysts.
• Programmatic report output. TradingAgentsGraph.save_reports() writes the same report tree the CLI produces, for headless and API runs. (#1037)
• Env-configurable reasoning depth via TRADINGAGENTS_OPENAI_REASONING_EFFORT, TRADINGAGENTS_GOOGLE_THINKING_LEVEL, and TRADINGAGENTS_ANTHROPIC_EFFORT, each gated to the models that accept it.

Changed

• Verified data-access contract. Symbol normalization on every vendor path (identity, returns, CLI, news); the configured vendor list is the exact resolution chain with no silent fallback to unselected vendors; a typed VendorError taxonomy; look-ahead-safe news windows; stale-OHLCV rejection; inclusive yfinance date ranges.
• Config precedence. An explicit TRADINGAGENTS_* value or CLI flag now wins over interactive defaults for debate and risk round counts, --checkpoint / --no-checkpoint, and the Docker provider profile; invalid boolean env values fail loudly. (#975, #976, #977)
• Current-generation model catalog. Refreshed provider lineups; retired gpt-4.1, Claude Sonnet 4.5, and the Gemini 2.5 line.
• Optional vendors degrade instead of aborting a run: a failed macro or prediction-market lookup returns a no-data sentinel.
• Analyst prompts lead with the current date so tool-call date ranges anchor to the run date rather than the model’s training cutoff. (#836)

Fixed

• Instrument identity. Deterministic ticker-to-company resolution prevents wrong-company hallucination, and a verified market-data snapshot grounds price and indicator claims. (#814, #830)
• Social and market data sources. Reddit RSS-first with 429 backoff, StockTwits transport hardening, and Alpha Vantage timeout plus key-versus-rate-limit handling.
• Structured output. Local OpenAI-compatible servers no longer reject object-form tool_choice; a thinking model that returns no parsed result falls back to free text; null-ish strings in optional price fields coerce to None. (#1038, #1051, #1057)

Removed

• The no-op analyst_concurrency_limit config knob; parallel analyst execution is planned for a later release. (#979)
• The unused committed uv.lock. (#1030)

Contributors

Thanks to everyone who shaped this release through code, design, and reports:

@CadeYu, @Zavianx, @weijianz-opc, @naltun, @brahmasky, @nik2208, @thieucong98, @Derekko-web, @LukiPrince, @Eddieargenal, @Ghraven, @ms32035, @yting27, @nyxst4ck, @KenCheung-AIxFinance, @yangyusheng2n, @fareloj, @haosenwang1018, @octo-patch, @seifenk, @CaoYuhaoCarl, @mihailnica10, @Dado-hash, @Handsomemikezzz, @ydhawesome, @macd2, @AyushKar2005, @wildhuman, @robert23kim, @bngness, @tedix-rodrigo, @malaccan, @rfalken78, @dengli1971-droid, @proofconcept39, @prasta1, @liximin, @jeffhuen, @mazar, @soyangelromero, @CNQQC, @dovetaill, @fperdigon, @gyx09212214-prog, @RSXLX.

[0.2.5] — 2026-05-11

Added

• Grounded Sentiment Analyst. The renamed sentiment_analyst now reads real Yahoo News, StockTwits, and Reddit data before generating its report, replacing the prior flow that could fabricate social posts under prompt pressure. (#557, #607)
• MiniMax provider with the full M2.x catalog (M2.7 / M2.5 / M2.1 / M2 plus highspeed variants, 204K context). Dual-region: Global (MINIMAX_API_KEY) and China (MINIMAX_CN_API_KEY).
• Dual-region Qwen and GLM with separate keys per region — international (DASHSCOPE_API_KEY, ZHIPU_API_KEY) and China (DASHSCOPE_CN_API_KEY, ZHIPU_CN_API_KEY), selectable via a secondary region prompt. (#758)
• TRADINGAGENTS_* env-var configurability for DEFAULT_CONFIG. Override llm_provider, deep/quick model IDs, backend_url, output_language, debate-round counts, checkpoint flag, and benchmark ticker via .env with type-aware coercion (string / int / bool). (#602)
• Interactive API-key detection in the CLI. When the selected provider’s key is missing, the CLI prompts for it and persists the value to .env so the analysis run continues without restart.
• Remote Ollama support. OLLAMA_BASE_URL points the CLI and the programmatic client at a remote ollama-serve. The CLI surfaces the resolved endpoint and warns on common malformed inputs. Adds a "Custom model ID" option for models pulled via ollama pull. (#648, #768)
• Configurable news-fetch parameters in DEFAULT_CONFIG — per-ticker article limit, macro headline limit, lookback window, and macro search queries. (#606, #683)
• Configurable alpha benchmark for non-US tickers. Replaces hardcoded SPY with regional indices for .NS (^NSEI), .T (^N225), .HK (^HSI), .L (^FTSE), .TO (^GSPTSE), .AX (^AXJO), .BO (^BSESN); explicit benchmark_ticker override available. Eliminates FX drift dominating alpha for non-USD listings. (#628, #684)
• Multi-language output covers every user-facing agent — researchers, risk debators, research manager, and trader, ending the previous partial-localization reports. (#575)
• Model catalog refresh. OpenAI GPT-5.5 frontier, Anthropic Claude Opus 4.7, Gemini 3.1 Flash-Lite GA, xAI Grok 4.20, Qwen 3.6 line. Versioned IDs only; auto-shifting aliases moved to the "Custom model ID" option.

Changed

• Sentiment Analyst is now consistently named across the CLI dropdown, status panel, and final reports (previously the backend was renamed but the CLI still said “Social Analyst”). The AnalystType.SOCIAL = "social" wire value is kept for saved-config back-compat.

Fixed

• Structured output works on DeepSeek V4 / reasoner and MiniMax M2.x. Those providers reject tool_choice per their tool-calling docs; the binding flow now skips it automatically via a capability table.
• pip install . installations pick up the project .env when running the CLI as a console script. (#747)
• Reports save end-to-end — streamed chunks were previously dropped from complete_report.md. (#719, #736)
• Ticker prompt preserves exchange suffixes (.SH, .SZ, .SS, .HK, .T, etc.) for A-share, HK, Tokyo, and other non-US flows. (#770)
• Docker permission errors no longer block first-run write to ~/.tradingagents/. (#519, #627, #672, #771)
• Config state no longer leaks between runs when sub-dicts are mutated; set_config partial updates preserve sibling defaults. (#788)
• max_recur_limit config actually applies — previously read but not forwarded to the propagator. (#764)
• Missing-API-key error names the exact env var to set. (#680)
• Quieter startup — suppressed the noisy upstream LangChainPendingDeprecationWarning from langgraph-checkpoint; will be removed once that package ships its fix.

Security

• Ticker path-traversal validation at every filesystem-path site (cache, checkpoint database, results) so a malicious ticker cannot escape its intended directory. (#618)

0.2.4 — 2026-04-25

Added

• Structured-output decision agents. Research Manager, Trader, and Portfolio Manager now use llm.with_structured_output(Schema) on their primary call and return typed Pydantic instances. Each provider’s native structured-output mode is used (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. (#434)
• LangGraph checkpoint resume — opt-in 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/. --clear-checkpoints resets them. (#594)
• Persistent decision log replacing the per-agent BM25 memory. Decisions are stored automatically at the end of propagate(); the next same-ticker run resolves prior pending entries with realised return, alpha vs SPY, and a one-paragraph reflection. Override path with TRADINGAGENTS_MEMORY_LOG_PATH. Optional memory_log_max_entries config caps resolved entries; pending entries are never pruned. (#578, #563, #564, #579)
• DeepSeek, Qwen (Alibaba DashScope), GLM (Zhipu), and Azure OpenAI providers, plus dynamic OpenRouter model selection.
• Docker support — multi-stage build with separate dev and runtime images.
• scripts/smoke_structured_output.py — diagnostic that exercises the three structured-output agents against any provider so contributors can verify their setup with one command.
• 5-tier rating scale (Buy / Overweight / Hold / Underweight / Sell) used consistently by Research Manager, Portfolio Manager, signal processor, and the memory log; Trader keeps 3-tier (Buy / Hold / Sell) since transaction direction is naturally ternary.
• Pytest fixtures — lazy LLM client imports plus placeholder API keys so the test suite runs cleanly without credentials. (#588)

Changed

• backend_url default is now None rather than the OpenAI URL. Each provider client falls back to its native default. The previous default leaked the OpenAI URL into non-OpenAI clients (e.g. Gemini), producing malformed request URLs for Python users who switched providers without overriding backend_url. The CLI flow is unaffected.
• All file I/O passes explicit encoding="utf-8" so Windows users no longer hit UnicodeEncodeError with the cp1252 default. (#543, #550, #576)
• Cache and log directories moved to ~/.tradingagents/ to resolve Docker permission issues. (#519)
• SignalProcessor reads the rating from the Portfolio Manager’s rendered markdown via a deterministic heuristic — no extra LLM call.
• OpenAI structured-output calls default to method="function_calling" to avoid noisy PydanticSerializationUnexpectedValue warnings emitted by langchain-openai’s Responses-API parse path. Same typed result, no warnings.

Fixed

• Empty memory no longer triggers fabricated past-lessons in agent prompts; the memory-log redesign makes this structurally impossible since only the Portfolio Manager consults memory and only when entries exist. (#572)
• Tool-call logging processes every chunk message, not just the last one, and memory score normalization handles empty score arrays. (#534, #531)

Removed

• FinancialSituationMemory (the per-agent BM25 system) and the dead reflect_and_remember() plumbing; subsumed by the persistent decision log.
• Hardcoded Google endpoint that caused 404 when langchain-google-genai changed its API path. (#493, #496)

Contributors

Thanks to everyone who shaped this release through code, design, and reports:

• @claytonbrown — checkpoint resume (#594), test fixtures (#588), design feedback on cost tracking (#582) and structured validation (#583)
• @Bcardo — memory-log redesign (#579), empty-memory hallucination report (#572), encoding fix proposal (#570)
• @voidborne-d — memory persistence design (#564), portfolio manager state fix (#503)
• @mannubaveja007 — structured-output feature request (#434)
• @kelder66 — RAM-only memory issue (#563)
• @Gujiassh — tool-call logging fix (#534), test stub PR (#533)
• @iuyup — memory score normalization fix (#531)
• @kaihg — Google base_url fix (#496)
• @32ryh98yfe — Gemini 404 report (#493)
• @uppb — OpenRouter dynamic model selection (#482)
• @guoz14 — OpenRouter limited-model report (#337)
• @samchenku — indicator name normalization (#490)
• @JasonOA888 — y_finance pandas import fix (#488)
• @tiffanychum — stale import cleanup (#499)
• @zaizou — Docker permission issue (#519)
• @Stosman123, @mauropuga, @hotwind2015 — Windows encoding bug reports (#543, #550, #576)
• @nnishad, @atharvajoshi01 — encoding fix proposals (#568, #549)

0.2.3 — 2026-03-29

Added

• Multi-language output for analyst reports and final decisions, with a CLI selector. Internal agent debate stays in English for reasoning quality. (#472)
• GPT-5.4 family models in the default catalog, with deep/quick model split.
• Unified model catalog as a single source of truth for CLI options and provider validation.

Changed

• base_url is forwarded to Google and Anthropic clients so corporate proxies work consistently across providers. (#427)
• Standardised the Google api_key parameter to the unified api_key form.

Fixed

• Backtesting fetchers no longer leak look-ahead data when curr_date is in the middle of a fetched window. (#475)
• Invalid indicator names from the LLM are caught at the tool boundary instead of crashing the run. (#429)
• yfinance news fetchers respect the same exponential-backoff retry as price fetchers. (#445)

Contributors

• @ahmedk20 — multi-language output (#472)
• @CadeYu — model catalog typing (#464)
• @javierdejesusda — unified Google API key parameter (#453)
• @voidborne-d — yfinance news retry (#445)
• @kostakost2 — look-ahead bias report (#475)
• @lu-zhengda — proxy/base_url support request (#427)
• @VamsiKrishna2021 — invalid indicator crash report (#429)

0.2.2 — 2026-03-22

Added

• Five-tier rating scale (Buy / Overweight / Hold / Underweight / Sell) introduced for the Portfolio Manager.
• Anthropic effort level support for Claude models.
• OpenAI Responses API path for native OpenAI models.

Changed

• risk_manager renamed to portfolio_manager to match the role description shown in the CLI display.
• Exchange-qualified tickers (e.g. 7203.T, BRK.B) preserved across all agent prompts and tool calls.
• Process-level UTF-8 default attempted for cross-platform consistency (note: this approach did not actually take effect; replaced in v0.2.4 with explicit per-call encoding="utf-8" arguments).

Fixed

• yfinance rate-limit errors are retried with exponential backoff. (#426)
• HTTP client SSL customisation is supported for environments that need custom certificate bundles. (#379)
• Report-section writes handle list-of-string content gracefully.

Contributors

• @CadeYu — exchange-qualified ticker preservation (#413)
• @yang1002378395-cmyk — HTTP client SSL customisation (#379)

0.2.1 — 2026-03-15

Security

• Patched langchain-core vulnerability (LangGrinch). (#335)
• Removed chainlit dependency affected by CVE-2026-22218.

Added

• pyproject.toml build-system configuration; the project now installs via modern packaging tooling.

Removed

• setup.py — dependencies consolidated to pyproject.toml.

Fixed

• Risk manager reads the correct fundamental report source. (#341)
• All open() calls receive an explicit UTF-8 encoding (initial pass).
• get_indicators tool handles comma-separated indicator names from the LLM. (#368)
• Propagation initialises every debate-state field so risk debaters never see missing keys.
• Stock data parsing tolerates malformed CSVs and NaN values.
• Conditional debate logic respects the configured round count. (#361)

Contributors

• @RinZ27 — langchain-core security patch (#335)
• @Ljx-007 — risk manager fundamental-report fix (#341)
• @makk9 — debate-rounds config issue (#361)

0.2.0 — 2026-02-04

This is the largest release since the initial public version. The framework moved from single-provider to a multi-provider architecture and grew several production-ready surfaces.

Added

• Multi-provider LLM support (OpenAI, Google, Anthropic, xAI, OpenRouter, Ollama) via a factory pattern, with provider-specific thinking configurations.
• Alpha Vantage integration as a configurable primary data provider, with yfinance as a community-stability fallback.
• Footer statistics in the CLI: real-time tracking of LLM calls, tool calls, and token usage via LangChain callbacks.
• Post-analysis report saving — the framework writes per-section markdown files (analyst reports, debate transcripts, final decision) when a run completes.
• Announcements panel — fetches updates from api.tauric.ai/v1/announcements for the CLI welcome screen.
• Tool fallbacks so a single vendor outage does not stop the pipeline.

Changed

• Risky / Safe risk debaters renamed to Aggressive / Conservative for consistency with the displayed agent labels.
• Default data vendor switched to balance reliability and quota across community deployments.
• Ollama and OpenRouter model lists updated; default endpoints clarified.

Fixed

• Analyst status tracking and message deduplication in the live display.
• Infinite-loop guard in the agent loop; reflection and logging hardened.
• Various data-vendor implementation bugs and tool-signature mismatches.

Contributors

This release is the first with substantial outside contributions; many community PRs from late 2025 also landed here.

• @luohy15 — Alpha Vantage data-vendor integration (#235)
• @EdwardoSunny — yfinance fetching optimisations (#245)
• @Mirza-Samad-Ahmed-Baig — infinite-loop guard, reflection, and logging fixes (#89)
• @ZeroAct — saved results path support (#29)
• @Zhongyi-Lu — .env gitignore (#49)
• @csoboy — local Ollama setup (#53)
• @chauhang — initial Docker support attempt (#47, later reverted; the merged Docker support shipped in v0.2.4)

0.1.1 — 2025-06-07

Removed

• Static site assets that had been bundled with v0.1.0; the public site now lives separately.

0.1.0 — 2025-06-05

Added

• Initial public release of the TradingAgents multi-agent trading framework: market / sentiment / news / fundamentals analysts; bull and bear researchers; trader; aggressive, conservative, and neutral risk debaters; portfolio manager. LangGraph orchestration, yfinance data, per-agent BM25 memory, single-provider OpenAI integration, interactive CLI.

This site is open source. Improve this page.