Financial Evidence for Python and AI agents

Use Seiche, LiquiLens, Undertow and Palimpsest financial research from Python, the terminal or an MCP client. Review funding conditions, bank disclosures, market liquidity and economic source availability with original sources, dates and coverage gaps attached. Operated by LIQUILENS PRIVATE LIMITED.

Try the research desk · First-review guide · Connect your tools

Install and save your first review

Python 3.10 or newer. The base package has no required dependencies; this review needs no model, account or data API key:

python3 -m pip install --upgrade financial-evidence
financial-evidence start
financial-evidence review funding --output reviews

start shows the available tasks offline. review makes one bounded public research request and saves the original result.json, a readable brief.md and a hash receipt. Inspect source dates and gaps before using the evidence. At your next review, compare against the saved capture:

financial-evidence review funding --output reviews \
  --previous reviews/YOUR-PREVIOUS-CAPTURE/result.json

Other starting tasks include covered institutions and indicative BTC exit costs:

financial-evidence review institutions --selection au-sfb --output reviews
financial-evidence review exit --selection 10000 --output reviews

Use --verification for installation or CI checks so those reads are labelled synthetic. Installation starts no scheduler. An unavailable result remains a gap.

Retain reviews when a source becomes unavailable

Version 0.1.14 preserves access-denial and rate-limit diagnostics, and adds local-only workflow show and workflow compare commands for saved runs:

financial-evidence workflow show RUN_ID --state ./funding-reviews --format markdown
financial-evidence workflow compare EARLIER_RUN_ID LATER_RUN_ID --state ./funding-reviews --format json

These commands read retained evidence without requesting fresh source data. Original observation clocks and hashes remain attached. An unavailable or blocked source remains a gap; saved output is not a current market quote. Learning and remote sharing remain explicitly opt-in.

Choose the product for your question

Coverage and permitted use differ by source. Research results do not grant trade execution, credit approval or redistribution rights; BTC estimates are not executable quotes.

Use Python or connect an AI agent

The same base package can save a review from your application:

from financial_evidence.research import fetch_workflow, save_review

result = fetch_workflow("funding", "USD")
capture = save_review(result, "reviews")
print(result["prepared_response"], capture / "brief.md")

For an MCP client, use the hosted Streamable HTTP endpoint:

https://agents.seiche.info/products/financial-evidence/mcp

Start with financial_evidence_agent_query and {"dataset":"money_markets","entity":"USD","limit":3}. Download the product MCP configurations to connect individual products; each has its own tools and scope.

For the dependency-free local stdio router:

uvx --from financial-evidence financial-evidence mcp

If already installed, run financial-evidence mcp. The local router exposes financial_evidence_topics, financial_evidence_route and financial_evidence_fetch; the hosted Research Desk has a separate tool set.

Repeat reviews in your application

The package includes AgentWorkflow and the workflow CLI, with a private local journal and comparisons against the last successful capture. On Linux or macOS:

financial-evidence workflow run funding --selection USD --state ./funding-reviews
financial-evidence workflow report --state ./funding-reviews
financial-evidence workflow history --state ./funding-reviews

This path also needs no optional dependencies. Your application chooses when to run it; installation does not enable polling or remote learning submissions. Use --mode test on workflow run for installation checks. See the agent workflow guide for Python hooks, monitor checks, feedback and explicitly enabled sharing.

Optional persistent MCP sessions and source-status streams

Install the workspace extra for persistent MCP/HTTP clients:

python3 -m pip install --upgrade 'financial-evidence[workspace]'
financial-evidence-trading doctor
financial-evidence-trading context

import asyncio
from financial_evidence.trading import ProductClient

async def main():
    async with ProductClient(mode="test", timeout=15) as client:
        first = await client.call("seiche", "data_health")
        again = await client.call("seiche", "data_health")
        print(first["timing"], again["timing"])

asyncio.run(main())

mode="test" labels verification traffic. Sessions are bounded and failed calls are not automatically replayed. Source-status snapshots and SSE events report metadata, not live quotes; session reuse cannot make sources fresher. Follow the persistent client and stream guide for expiry and reconnect handling. The separate openbb extra supports the OpenBB integration.

Source · Issues · Privacy · Licence