NORTHPOINT
A signed layer of provenance, lineage, and source reliability — the data trust layer on which autonomous agents operate with full traceability.
Built for finance teams who need every claim grounded, every action audit-defensible, and a compliance officer who can sign off on the answer.
The morning inbox shows what reconciled cleanly, what didn’t, which sources are degrading, and which agent proposals are waiting on you — all linked to signed lineage rows.
An LLM acting on unverified data is a regulatory liability. Northpoint stacks autonomous agents on top of a signed trust layer — so every answer points to evidence a compliance officer can sign off on.
Break investigator · Corp actions · Anomaly explainer · Ops triage. Each action grounded in a signed lineage chain.
Per-column profiles, dataset semantic index, lineage trace API, per-org LLM credentials, tool catalogue exposed over MCP.
Lineage + golden values, SHA-256 hash chain, source reliability tiers, deterministic data-quality engines, connector cache fallback.
LLMs propose. Deterministic engines enforce. Every proposal lands as a reviewable artifact — never a direct mutation. The audit chain only records deterministic actions.
Four copilots, each thin layered over an existing surface in the product. Each one solves a distinct compliance-shaped pain — and never bypasses the operator.
Operator gets a finding. Then walks lineage manually.
Reads the field's last N observations, the column's profile, and recent similar incidents. Drafts a 2–3 sentence operator-language explanation with a lineage citation on every numeric claim.
Yahoo reported qty=12,500 for AAPL — 8.3× the column's typical value. Alpaca and IBKR both reported 1,500. Likely a Yahoo decimal-place error (similar issue resolved 2026-04-12). Recommended: ignore Yahoo for this row.
A position break fires. Hours of source-by-source archaeology.
Walks the lineage chain backward, cross-references the reliability tier and uptime history of each source, and outputs a signed evidence packet auditors can replay.
Alpaca file arrived 2h late on 2026-05-20; the recon ran against stale cached bytes. Packet signed with the chain hash at investigation time.
Splits, dividends, mergers — adjusted across brokers, errors silent until next recon.
Ingests CA feeds, drafts the adjustment per account, and surfaces conflicts (divergent record dates, missing positions, tax-lot differences) as blocking approvals.
Adjustment queued for AAPL 4-for-1 split: IBKR 1,200 → 4,800; Alpaca 800 → 3,200. Conflict: divergent record dates between sources — operator review required.
Ops questions that can't be answered through point-and-click.
Natural-language question bar over catalog + lineage + incidents + findings + runbook history. Every numeric answer has a 'show evidence' expand revealing the lineage refs and the RAG query.
Q: Which datasets had the most finding-volume this week? A: prices.alpaca, prices.yahoo, recon.daily — citing 38, 31, 24 findings respectively.
Cross-source reconciliation, source-reliability tiers, the operator inbox, the signed audit trail — the load-bearing primitives the AI roadmap sits on top of.
Broker vs internal, market vs mark, expected vs arrived — reconciled across every source, with an audit trail that proves what fired, who saw it, and how it was resolved.
Mismatches, freshness breaches, and source-health regressions roll up into one feed. Notified once when something starts failing and once when it recovers — not on every tick.
Brokers, market data, on-chain inputs, spreadsheets, internal files — connected once and tracked for time-weighted uptime, so you know which feeds you can trust before you act on what they say.
Reconciliation runs, incident acks, resolutions, ack-by-whom and ack-when, root cause, downstream notifications — logged, timestamped, and exportable for review.
Reconciled inputs, accountable controls, and checks that run on schedule — the way data trust compounds.
Brokers, market feeds, spreadsheets, webhook streams, internal files — cross-checked against each other so what reaches your desk has already been agreed by every source that touches it.
Role-based access, audit trail, ack/resolve state, root-cause notes — the controls that turn alerts into accountable incidents.
Saved checks run on schedule, runbooks codify the response, source-health scores trend over time — so trust gets built, not re-litigated every morning.
Workflow review, Northpoint deployed against your stack, full documentation, and continued support — so the trust layer outlasts the engagement.
01
Map current tools, manual checks, alert paths, and reporting requirements.
02
Connect the data, ship the operating surface, configure alerts, and document the system.
Practical answers ahead of a short workflow review — what Northpoint is, who it’s for, and how it lands on your stack.
Northpoint is the agentic data trust layer for finance: a signed layer of provenance, lineage, and source reliability — with autonomous copilots that operate on top with full traceability. Every claim is grounded in a hash-chained lineage row, so compliance can sign off on the answer, not just the output.
Finance teams whose work is regulated, audited, or compliance-reviewed: trading desks, asset managers, RIAs, family offices, wealth teams, and ops desks at funds whose data-quality requirements have outgrown spreadsheets and BI dashboards.
LLMs propose. Deterministic engines enforce. Every AI proposal that touches operator state lands as a reviewable artifact — a rule, an alias mapping, a draft runbook, a finding narrative — never a direct mutation. The audit chain only records deterministic actions; LLM calls are logged separately as advisory traces.
Only if you choose. PII and cell-values are redacted before any LLM send; raw rows require explicit per-call consent. Customers route through their own Anthropic / OpenAI / Bedrock / Vertex / Azure OpenAI key — Northpoint never proxies your bytes through our infrastructure unless you opt in. An on-device embedding option exists for high-security customers.
Northpoint is a product, tuned and deployed against your brokers, feeds, files, and internal systems. A focused dashboard sprint ships in 7–14 days; a full operating-system build with the trust layer and the copilot suite is 3–5 weeks depending on integrations.
Yes. Typical inputs include IBKR / Alpaca / broker exports, market data feeds, TradingView, Polygon, Yahoo, Google Sheets, Notion, Slack, Telegram, Discord, and internal files. Brokers are treated as canonical; market-data sources are explicit opt-in fallbacks.