The Financial Intelligence API your product can build on
Builders want grounded financial intelligence they can call, not a model they have to babysit. LyraMind exposes a small set of versioned primitives, a readiness gate agents check before they act, and an MCP server, with a Trust Layer on every response.
The problem: raw models are ungrounded and unauditable
Wiring an LLM to financial questions gives you fluent answers with no grounding, no sources, and no record of what they were built from. For a product, that is a liability: you cannot show a user where an answer came from, and you cannot let an agent act on it safely.
What builders need is an API that returns grounded reads with citations and confidence, plus a gate an agent can check before it does anything consequential.
What LyraMind gives you: primitives, a gate, and MCP
The versioned Intelligence API (mounted at /v1) is built on three primitives every client calls: explain (audience-tailored read of a symbol), company (structured read), and summarize (condense text or a symbol read). On top of those sit readiness (an AGENT_READY / REVIEW_REQUIRED / DO_NOT_EXECUTE gate an agent checks before it executes), verify (multi-model consensus), committee, pulse, graph, twin, and an agent marketplace.
Execution stays pluggable and orchestrated, never owned: the execution-readiness decision reports whether an action is ready and which connected providers could execute it, but LyraMind never moves funds, custodies assets, or trades. Every response carries the Trust Layer, and self-serve onboarding issues an org, an owner, and a first test API key, with tiered rate limits and per-org metering.
The MCP server exposes these as tools (lyramind_explain, lyramind_company, lyramind_score, lyramind_agent_readiness, lyramind_execution_readiness, lyramind_disclosure, lyramind_ledger_audit, and more) so Claude, ChatGPT, and other agent hosts can call them directly. The public demo runs on sample data; live news needs a provider key.
How you use it
Self-serve onboard to get an org and a test key, call the three primitives for grounded reads, and put the readiness gate in front of any consequential action so your agents never act on unverified intelligence. Drop the MCP server into your agent host to expose the same capability as tools.
Rate limits and metering are per-org from the start, and every read records to the Trust Ledger for auditability.
- Three versioned primitives: explain, company, summarize (GET and POST)
- Agent Readiness gate: AGENT_READY / REVIEW_REQUIRED / DO_NOT_EXECUTE before an action
- Multi-model verification, committee, pulse, graph, twin, and agent marketplace endpoints
- MCP server exposing the engine as tools for Claude, ChatGPT, and other hosts
- Self-serve onboarding: org, owner, and first API key, with per-org metering and rate limits
- Trust Layer on every response and a Trust Ledger for auditability
Questions
Can LyraMind execute trades or move money for my app?
No. LyraMind is the intelligence and decision layer. It decides whether an action is ready and which connected providers could execute it, but it never moves funds, custodies assets, or trades. Execution stays pluggable.
How do responses stay auditable?
Every response carries the Trust Layer (confidence, sources, evidence, reasoning, data-mode, freshness) and every read records to the Trust Ledger, so you can show users where an answer came from.
How do I get started and manage keys?
The accounts onboarding endpoint self-serves an org, an owner, and a first test key. From there you can create, rotate, and revoke keys, and usage is metered per org against tiered rate limits.