Financial evidence for agents, risk teams and researchers.
Put the latest upgrades into one useful task: a cited funding check, an institution watchlist or a review of market exit costs. Keep the result and return to the same question tomorrow.
New agent infrastructure 1.1.0: persistent Python MCP sessions, a connection doctor and Python/JavaScript source-status clients. Free public research reads; source metadata and evidence retain their own dates and restrictions.
Run these three workflows in your browser, save a baseline and compare the next review.
Public research needs no data API key. Copy the task instructions · Machine-readable workflows. Interface checks: 9 October 2026. Each observation retains its own source date.
Finance-agent developers: get a cited funding result
Add one named dataset to your existing research step. This request returns a bounded USD funding sample, its original observation date, source and coverage:
curl --fail --max-time 30 \ 'https://agents.seiche.info/v1/research/api/v1/agent-query?dataset=money_markets&entity=USD&limit=5'
For MCP, connect https://agents.seiche.info/products/financial-evidence/mcp, discover its schemas and call financial_evidence_agent_query with {"dataset":"money_markets","entity":"USD","limit":5}. Inspect next_offset for more rows. Save the response and its revision; repeat the same task on another day, checking source dates even when the result is unchanged.
Run and save a funding watch · Copy the agent instruction · REST documentation
Treasury and bank-risk teams: start with your watchlist
The new LiquiLens institution monitor organizes supported deterioration alerts, source-dated disclosures and evidence gaps. Choose exact institution identifiers from the live register:
curl --fail --max-time 30 \ 'https://agents.seiche.info/v1/data/liquilens.institution_watchlist?slugs=au-sfb,bajaj-finance'
Or call institution_evidence_monitor at https://agents.seiche.info/products/liquilens/mcp with {"slugs":["au-sfb","bajaj-finance"]}. These are illustrative institutions. Retain the review queue for your own counterparties and compare the next working day's disclosures and unresolved gaps. Missing evidence does not establish safety; a disappearing warning is not proof of recovery.
Open institution monitoring · Copy the watchlist instruction
Traders and researchers: review funding and exit evidence
Check Seiche data_health and Undertow liquidity_tiers first. Then use these tools through the direct connections below:
Seiche money_market_context:
{"section":"summary"}
Undertow crypto_exit_check:
{"sizes_usd":[10000,100000],"max_age_seconds":7200}Keep funding conditions and hypothetical BTC sell-size estimates in separate sections of your note. Preserve source dates, venue coverage and withheld values; fees are excluded and estimates are not executable quotes. Repeat the same sizes on the next research day.
Open funding research · Open the exit desk · Copy the research instruction
Keep the workflow useful
Retain the first output, record what it helped you review, and return for a second task. Use the 14-day evaluation scorecard to track usefulness and blockers with the participant's consent. The developer, risk-team and individual-researcher cohorts stay separate. These examples do not enroll users or configure background jobs.
Once a first task works, set up recurring research with retained receipts and recovery. NoiseFloor 0.4.0 adds correlation research and descriptive analysis of your own entitled series and headlines. Open the correlation workbench. The browser paper walkthrough separately rehearses order controls with synthetic data.
Check a trading-agent integration
Run the connection readiness check to verify public MCP sessions and inspect current source restrictions. Use the updated private paper-host guide for exact assessments and customer-authorized outcome sharing.
Connect each product directly
Choose LiquiLens for covered-bank evidence, Seiche for funding context, and Undertow for market-liquidity research. Download the portable MCP configuration or VS Code configuration and merge only the entries you need into your existing settings. Read the connection guide · Machine-readable connection catalog.
| Product | Start with | Access and coverage |
|---|---|---|
| LiquiLens | institution_research_coverage | Public research on covered institutions and disclosures |
| Seiche | data_health | Source-specific funding observations, clocks and rights |
| Undertow | liquidity_tiers | Public summaries; subscriber and full-fidelity tools keep their own entitlements |
The catalog records a dated endpoint and tool-inventory check. It does not establish current source data or test every tool. These connections provide research evidence and cannot place orders or transfer funds.
Use your existing framework
python3 -m pip install --force-reinstall 'financial-evidence @ git+https://github.com/beepboop2025/financial-evidence-skills.git@trading-readiness-v1.0.0' langchain-core
from financial_evidence.agents import EvidenceAgentClient, framework_tools
with EvidenceAgentClient() as evidence:
tools = framework_tools('langchain', evidence)
print(tools[0].invoke({})) # Offline catalog.
# Attach tools to your researcher model and its execution node.Native adapters support LangChain/LangGraph, CrewAI, OpenAI Agents and Pydantic AI. Plain typed Python functions work with custom researchers and schedulers. Install your chosen SDK alongside the package; your model provider has its own requirements.
Repeat checks without repeating every row
Save the response's revision and pass it as previous_revision on the same query. Identical published evidence omits duplicate rows while returning source metadata. Partial and failed retrievals always return diagnostics. Unchanged does not mean fresh. Check source dates, availability, rights and limitations on every run.
Connect an institutional research pipeline
| Workflow | Useful input | Evidence boundary |
|---|---|---|
| TradingAgents, FinRobot, custom research agents | Funding, covered-bank and liquidity research tools | Complement existing price and fundamentals providers |
| Qlib, FinRL, LEAN, NautilusTrader, vectorbt | Cited captures for exploratory feature research | Validate knowledge time and vintages before backtesting |
| GS Quant, ArcticDB, OpenBB | Dated research tables and receipts | Integration recipes do not imply institutional adoption |
Export a table with the capture example. It preserves rows and a SHA-256 receipt in a new directory. History filters select currently published observations; they do not establish what was knowable in the past.
Know the coverage
Each value retains its original date, unit, source URL, source path and content hash. Withheld or unavailable values remain missing. Bank diagnostics are research measures; market percentiles are not executable quotes. These tools supply evidence for human-reviewed research and have no trade-execution authority.
Read framework setup and quant pipeline recipes · Repository integration map · Registration demo