Surface live, market-implied probabilities for forward-looking events (Fed
decisions, recession, elections, geopolitics, crypto) to the news analyst via a
new get_prediction_markets tool and a prediction_markets vendor category. Backed
by Polymarket's public Gamma API (no key). Results are filtered to open,
forward-looking markets (closed and past-dated events excluded), ranked by
traded volume, and rendered with implied probability, volume, resolution date,
and the recent move. External errors degrade to a clear unavailable message
rather than interrupting the analyst.
Surface Federal Reserve Economic Data (rates, inflation, labor, growth) to the
news analyst via a new get_macro_indicators tool and a macro_data vendor
category. Friendly aliases (cpi, unemployment, fed_funds_rate, 10y_treasury,
yield_curve, ...) map to FRED series IDs; raw series IDs are accepted too. The
report gives the latest value, change over the window, and a recent observation
table. Windowing is lookahead-safe (observation_end = curr_date), missing values
are skipped, and a missing FRED_API_KEY surfaces as a clear not-configured
condition through the vendor router rather than a crash.
The market analyst is bound to call get_verified_market_snapshot and its prompt
requires it as the source of truth, but the tool was missing from the market
ToolNode executor — so the call failed and the model reported it "unavailable"
and skipped verification. Register it (with a regression guard) so the snapshot
actually runs and grounds the report.
resolve_instrument_identity and the reflection return lookup queried Yahoo with
the raw ticker, so broker/forex/commodity symbols (XAUUSD, BTCUSD, EURUSD)
failed identity or could mismatch the priced instrument even though the price
path already normalized them. Route both through normalize_symbol (#983, #984).
Adds a cross-provider temperature config (and TRADINGAGENTS_TEMPERATURE),
forwarded to every LLM client when set, so runs can be made less variable
on models that honor it. Adds a README "Reproducibility" section that
separates the sources of run-to-run variation, what users can control
(temperature, non-reasoning model, pinned date), and what is inherent to
LLM-driven analysis, and notes that the identity and verified-data fixes
already removed the "different companies / fabricated prices" variance.
#178#168
Agents had no ground-truth ticker→company mapping, so the market analyst
could pattern-match a price chart to the wrong company (e.g. TOTDY read as
"TotalEnergies"), and every downstream agent inherited the bad framing.
Resolve identity once at run start via a cached, fail-open yfinance lookup
and inject company/sector/exchange into the shared instrument context that
all twelve agents consume, with an explicit do-not-substitute instruction.
Resolution runs on both the propagate() and CLI entry points.
Also replaces the bare "Continue" message-clear placeholder, which some
OpenAI-compatible providers interpreted as the user task, with a
context-anchored placeholder carrying the resolved identity and date.
#814#888
- analyst_execution.py: rename "Social Analyst" / "Msg Clear Social"
to "Sentiment Analyst" / "Msg Clear Sentiment" to match v0.2.5.
- conditional_logic.should_continue_social returns the renamed route.
- TradingAgentsGraph.propagate accepts asset_type and threads through
to Propagator.create_initial_state.
- Regression test on the Sentiment Analyst label.
Verified end-to-end (NVDA stock + BTC-USD crypto) on gpt-5.4-mini.
SPY was hardcoded as the alpha benchmark in both the return-fetch
path and the reflection label, which produced meaningless alpha for
.NS / .T / .HK / .L / .TO / .AX / .BO listings — FX drift between a
local-currency stock and a USD index dominates the spread.
DEFAULT_CONFIG now exposes benchmark_ticker (explicit override) and
benchmark_map (suffix → regional index, with SPY as the empty-suffix
default). TRADINGAGENTS_BENCHMARK_TICKER joins the env-overlay table.
Trading graph resolves the benchmark once per ticker and threads it
through to both _fetch_returns and reflect_on_final_decision, so the
alpha label reads "Alpha vs ^N225" for Tokyo listings, "Alpha vs ^HSI"
for Hong Kong, etc., instead of the misleading "Alpha vs SPY".
- dataflows/config: deepcopy + one-level dict merge so a partial
set_config doesn't clobber sibling defaults
- graph: thread max_recur_limit from config to Propagator
- openai_client: name the missing env var in the API-key error
#788#764#680
graph.stream() yields per-node deltas, not the full state. Taking
trace[-1] only captured the last node's contribution, so reports
saved to disk were missing every section except the final decision.
Merge all chunks in both the CLI path and trading_graph._run_graph's
debug branch.
#719#736
The ticker symbol reaches three filesystem-path construction sites
(load_ohlcv cache filename, checkpointer DB path, _log_state results
directory) without validation. A value containing path separators or
"../" escapes the configured cache / checkpoints / results directory.
Two attack vectors:
- Programmatic callers passing arbitrary ticker to propagate()
- Prompt injection via fetched news content steering the LLM into
tool calls with attacker-chosen ticker
Fix: new safe_ticker_component() validator in tradingagents/dataflows/
utils.py applied at all three sites. Allows the standard ticker
character set ([A-Za-z0-9._\-\^], up to 32 chars) and explicitly
rejects dot-only values like "." and ".." which would otherwise pass
the regex but traverse parent directories. Seven test cases cover
the accepted formats (BRK-B, 7203.T, ^GSPC, etc.) and the rejected
inputs (path separators, null bytes, whitespace, empty values,
overlong strings, dot-only values).
Closes#618.
Long analyses can take many minutes; a crash or interruption forced users
to re-run from scratch and re-pay every LLM call. This adds an opt-in
checkpoint layer backed by per-ticker SQLite databases so the graph
resumes from the last successful node.
How to use:
- CLI: tradingagents analyze --checkpoint
- CLI: tradingagents analyze --clear-checkpoints
- Python: config["checkpoint_enabled"] = True
Lifecycle:
- propagate() recompiles the graph with a SqliteSaver when enabled and
injects a deterministic thread_id derived from ticker+date so the
same ticker+date resumes while a different date starts fresh.
- On successful completion the per-thread checkpoint rows are cleared.
- The context manager is closed in a try/finally so a crash never
leaks the SQLite connection or leaves the graph in checkpoint mode.
Storage: ~/.tradingagents/cache/checkpoints/<TICKER>.db
(override via TRADINGAGENTS_CACHE_DIR).
The checkpointer module is new (tradingagents/graph/checkpointer.py)
and the GraphSetup now returns the uncompiled workflow so it can be
recompiled with a saver when needed.
Adds langgraph-checkpoint-sqlite>=2.0.0 dependency. 3 new tests verify
the crash/resume cycle and that a different date starts fresh.
The previous per-agent BM25 memory was effectively dead code — its only
caller was a commented-out line in main.py. Replace it with a single
append-only markdown decision log driven by the propagate() lifecycle.
Lifecycle:
- store_decision() appends a pending entry at the end of every run
- _resolve_pending_entries() runs at the start of the next same-ticker
run, fetches yfinance returns + alpha vs SPY, and writes one LLM
reflection per resolved entry through an atomic temp-file rename
- Portfolio Manager consumes state["past_context"] (5 most recent
same-ticker entries plus 3 cross-ticker reflection-only excerpts)
Storage at ~/.tradingagents/memory/trading_memory.md
(override: TRADINGAGENTS_MEMORY_LOG_PATH).
Tag schema:
- Pending: [YYYY-MM-DD | TICKER | Rating | pending]
- Resolved: [YYYY-MM-DD | TICKER | Rating | +X.X% | +Y.Y% | Nd]
Removes rank-bm25 dependency and the legacy reflect_and_remember()
plumbing across reflection.py, trading_graph.py, and the agent factories.
49 new tests in tests/test_memory_log.py cover the storage, deferred
reflection, prompt injection, and legacy-removal paths. Full suite
(58 tests) passes in under 2 seconds without API keys.
Add effort parameter (high/medium/low) for Claude 4.5+ and 4.6 models,
consistent with OpenAI reasoning_effort and Google thinking_level.
Also add content normalization for Anthropic responses.
Prevents UnicodeEncodeError on Windows where the default encoding
(cp1252/gbk) cannot handle Unicode characters in LLM output.
Closes#77, closes#114, closes#126, closes#215, closes#332
- Add StatsCallbackHandler for tracking LLM calls, tool calls, and tokens
- Integrate callbacks into TradingAgentsGraph and all LLM clients
- Dynamic agent/report counts based on selected analysts
- Fix report completion counting (tied to agent completion)
Yijia-Xiao