Metal
Metal is an AI context platform for private equity and private credit firms. It structures deals, documents and decisions into one Context Graph, then serves that context to Claude, ChatGPT or in-house agents.
What is Metal?
Metal is an AI context platform published by Metal Technologies Inc., a New York company founded by Taylor Lowe, Sergio Prada and James O'Dwyer. It addresses one problem: inside a private capital firm, what the firm knows sits scattered across a CRM, file repositories, data rooms, mailboxes and third-party data providers, and no AI assistant reasons well over that mess.
The answer is the Context Graph. Metal pulls those sources in and resolves them into companies, deals, people, documents, lists and workflows, linked by how they actually relate. Duplicates collapse, so that Apex Holdings Inc. and APEX Holdings LLC become one canonical record, and the graph updates continuously as new files arrive.
Two properties do the real work. The first is reconciliation: when the same figure appears in five documents with five values, Metal ranks them by document authority, source and recency, designates a canonical value and shows a confidence score. The second is provenance, since every value traces back to the exact document and page it came from. Metal also records decisions rather than documents alone, keeping the thesis, the risks flagged and the call that was made.
Metal argues this beats pointing an assistant straight at raw files through MCP, where structure is re-derived on every prompt, tool calls multiply, answers drift between colleagues and token costs climb. Structure once, answer indefinitely.
Crucially it does not replace the AI a team already uses. It sits between systems of record and those tools, exposed through an open REST API at api.metal.ai and a hosted MCP server at mcp.metal.ai carrying 45 tools, documented for Cursor, Claude and ChatGPT. Source permissions are inherited, so a user or an agent only ever sees what they should.
Three products sit on the graph, Context Graph, Workflows and Deal Intelligence, addressed to private equity and private credit. Security claims include SOC 2 Type II, ISO 27001 in progress, zero-trust architecture, AES-256 and TLS 1.3, SSO, role-based access and audit logs, alongside a repeated commitment never to train models on customer data. Onboarding is a guided project: read-only connections, first answers in days, full value quoted at 30 to 60 days.
What it does
- Structure a firm's CRM, file repositories, data rooms and mailboxes into a single Context Graph
- Reconcile the same metric across conflicting sources and rank one canonical value by document authority, source and recency
- Trace every figure back to the exact document and page it came from
- Answer questions about the firm from Claude, ChatGPT, Copilot or an in-house agent through a native MCP server
- Run agent workflows across the deal cycle, from business development one-pagers and data room Q&A to comparable deal benchmarks and IC memos
- Track the deal pipeline from sourcing to exit in one live view shared by the whole team
- Extract terms from term sheets, spread financials and monitor covenant headroom on the private credit side
When to use Metal / When not to
A quick filter to help you decide if Metal is the right fit.
When to use Metal
- Private equity deal teams running the full cycle from sourcing and screening through diligence, investment committee, portfolio monitoring and exit
- Private credit teams extracting terms from term sheets, spreading financials, drafting credit memos and tracking covenant headroom
- Technology, data and AI teams inside a fund who want to build on an open REST API, a native MCP server and event webhooks
- Fund leadership looking for one reconciled source of truth firm-wide, rather than answers that differ from one analyst to the next
- Firms whose knowledge is already scattered across a CRM, SharePoint or Box repositories, data rooms and Exchange mailboxes
When not to use Metal
- Individuals and small teams: there is no free plan, no free trial and no self-service sign-up, and the only way in is a sales demo
- Anyone working outside private markets, since the product is built exclusively around private equity and private credit workflows
- Buyers who need published pricing before a conversation, as Metal discloses no rate, tier or range anywhere on its site
- Mobile-first users: no iOS or Android application exists, and the product lives in a web app, an API and an MCP server
- Teams needing value within days without running a project, since Metal itself frames full value at 30 to 60 days after connecting sources
How to use Metal
A typical end-to-end flow, from setup to results.
- Request a demo through the form at the Book a Demo page, as there is no self-service sign-up
- Let Metal connect your sources: the team handles the connectors, clears duplicates and stale records and structures the data, over read-only connections
- Have an administrator enable firm-wide integrations under Settings then Integrations, browsing the available connections and installing the ones you need
- Ask each user to connect their personal accounts under Settings then Connections; Microsoft Exchange is per-user only and needs no admin step
- Configure the workspace: users, roles and permissions, taxonomy, custom attributes, deal profiles, the NDA library and dashboards
- Work in the app across companies, deals, people, documents and lists, using search, enrichment and dashboards
- To use MCP, have an admin switch on access to the Metal MCP server under Settings, Organization, API and MCP Access
- Add the hosted MCP server URL as a remote server in Cursor, Claude or ChatGPT and sign in through OAuth, choosing your organisation if you belong to more than one
- For programmatic access, create an API key and call the Metal API gateway, where endpoints are versioned under a /v1 prefix and return JSON
- Run the out-of-the-box workflows or compose your own from triggers, steps and agents, keeping human review wherever judgement matters
Pros & Cons
Pros
- Systematic provenance: every figure traces to a specific document and page, which is what makes the output auditable
- Conflicting values are reconciled explicitly, ranked and scored, rather than silently averaged or picked at random
- It layers onto the AI stack a firm already has, whether Claude, ChatGPT, Copilot or in-house agents, instead of replacing it
- Open standards throughout, with MCP, a REST API and customer-controlled identifiers backing an explicit no-lock-in claim
- A clear and repeated commitment never to train models on customer data, stated on three separate pages
- SOC 2 Type II obtained, ISO 27001 listed as in progress, and a public Trust Center
- Source-system permissions are inherited, data is logically isolated per firm, and everything is exportable and deletable on exit
Cons
- No public pricing whatsoever: no pricing page, no tiers, no range
- No free plan and no free trial, the only entry point being a sales demo
- The published terms are Website Terms of Use covering the site and its subdomains, not the SaaS agreement, so the real customer contract stays private
- No subprocessor list and no Data Processing Agreement on the public pages, and the Trust Center is JavaScript-rendered and unreadable without a browser
- No documented data residency, since the privacy policy permits storage anywhere in the world
- Narrow scope, private equity and private credit only, with no mobile application
- Time to value is quoted at 30 to 60 days, which makes this a project rather than a purchase
Pricing & Plans
Metal publishes no pricing. There is no pricing page, no link to one in the navigation or the footer, and no amount, currency or billing unit anywhere on the site. Neither a permanent free plan nor a free trial is announced. The only commercial entry point is the Book a Demo form, so terms are set through a sales conversation. The privacy policy names Stripe, Inc. as a third-party payment processor, which confirms that billing exists even though its terms are not disclosed.
Data, GDPR & hosting
A consolidated view of how Metal handles your data.
GDPR overview
Metal states GDPR compliance plainly on both its homepage and its security page: full compliance with EU data protection requirements. Concrete measures are documented. Non-essential cookies are only set for EU and EEA visitors after consent given through a Cookie Manager, withdrawable at any time, and marketing relies on prior consent in the EU and EEA. Transfers outside the EEA, the United Kingdom and Switzerland to countries without an adequacy decision rely on the European Commission's Standard Contractual Clauses and the UK Addendum, with a copy available on request at privacy@metal.ai. The privacy policy took effect on 21 October 2025. Three gaps remain: no Article 27 EU representative is named, no data protection officer is designated, and no Data Processing Agreement appears on any public page.
Who owns the data?
Metal Technologies Inc. and its licensors own the site and all Metal Content. You keep ownership of any User Content you submit through the site, but submitting it grants Metal a worldwide, perpetual, non-exclusive, irrevocable, royalty-free and sublicensable licence to use, reproduce, modify, distribute, display and create derivative works from it for any purpose, including promoting Metal's own products. Feedback goes further still: all rights in it are assigned outright. Read the scope carefully, because these are Website Terms of Use covering the site and its subdomains, not the SaaS agreement. For the platform itself, Metal states that data is logically isolated per firm, never trained on, and fully exportable and deletable on exit.
Reuse rights
Under the website terms your right of use is limited, non-exclusive, non-transferable and revocable, and granted for informational purposes only. You may not copy, reproduce, distribute or create derivative works from Metal Content, reverse engineer the site, or point crawlers and scrapers at it without written permission. Your own firm data is treated differently. Metal says it is fully exportable and deletable when you leave, and the REST API returns the same core records as JSON, carrying your own externalId and externalReference values so records reconcile against a CRM or a data warehouse. Metal presents this as deliberate: context is exposed through standards-based interfaces so that a firm's data is not locked into one chat surface or one model provider.
Data retention & training
Hosting summary
Metal documents jurisdiction loosely. The privacy policy states that all information it processes may be transferred, processed and stored anywhere in the world, including but not limited to the United States and other countries, and warns visitors from the EEA, the United Kingdom and Switzerland that destination countries may have less stringent data protection laws. For transfers outside those regions to countries without an adequacy decision, Metal relies on the European Commission's Standard Contractual Clauses and the UK Addendum, and will supply a copy on request at privacy@metal.ai. No guaranteed hosting country or region is published and no data residency option is offered. Encryption is stated as AES-256 at rest and TLS 1.3 in transit, with logical isolation per firm. A DNS lookup on 24 August 2026 resolved the domain to an anycast node in Amazon's AS16509 seen from Amsterdam, which indicates CDN infrastructure rather than where customer data lives. The Vanta Trust Center advertises security reports but renders in JavaScript and could not be read.
Things to keep in mind
Risks and trade-offs to weigh before adopting Metal.
- Outsourcing judgement: the platform drafts IC memos, fit scores and benchmarks, and it is easy to let an agent's output stand in for the reasoning an investment decision requires
- Confidence scores shown beside values can lower a reader's guard on a source that still deserves challenge; Metal's own guardrail is that every output stays reviewable with citations
- Concentrating deals, IC memos and credit agreements in a single system moves a large part of the firm's risk surface onto the vendor
- The published terms cover the website rather than the service, so retention, subprocessing and liability commitments stay invisible until they are negotiated
- Content submitted through the site carries a perpetual, irrevocable and sublicensable licence to Metal, and feedback is assigned outright
- The website terms cap liability at 100 US dollars and impose binding arbitration seated in New York under New York law
- If the graph becomes the firm's only working memory, leaving grows expensive in practice even though export is promised
Setup & Integrations
Technical difficulty
Low effort for the customer, but not instant. There is no self-service sign-up: onboarding starts with a sales demo, after which Metal connects the sources, clears duplicates and structures the data itself, over read-only connections. Your side is mostly administrative, as an admin enables integrations, each user connects their own accounts, and MCP access is switched on in organisation settings before signing in through OAuth from Cursor, Claude or ChatGPT. Developers create an API key and call versioned endpoints. Metal quotes first answers in days and full value in 30 to 60 days.
Deployment
Integrations
Behind Metal
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What is Metal?
Do we have to give up Claude or ChatGPT to use it?
How much does Metal cost?
Is there a free plan or a free trial?
Does Metal train AI models on our data?
What certifications does Metal hold?
Which systems can we connect?
Is there an API?
How long does it take to get value?
What happens to our data if we leave?
Should you pick Metal?
Metal is a deliberately narrow product, and that is its strength. It does not try to be a general assistant: it builds one reconciled, permission-aware graph of what a private equity or private credit firm already knows, then hands that context to whatever AI the team is using. Three things genuinely set it apart. Values are reconciled across conflicting sources and ranked by document authority, figures stay traceable to an exact document and page, and decisions are captured alongside documents, so the firm's reasoning survives the deal that produced it.
The posture is infrastructure rather than application. An open REST API, a native MCP server and customer-controlled identifiers back the claim that context is not locked into one chat surface or one model provider. The signals of seriousness are real: SOC 2 Type II, a public Trust Center, named clients including Clearlake Capital, Berkshire Partners, Valesco Industries and Blue Wolf Capital, and 5 million dollars raised from Base10.
What Metal does not tell you matters just as much. There is no published pricing, no free plan and no free trial. The terms of service on the site cover the website and its subdomains, not the SaaS product, so the retention, subprocessing and liability commitments that would actually govern a firm's data are settled privately. No subprocessor list is published, no Data Processing Agreement is offered publicly, and the privacy policy permits storage anywhere in the world.
For a fund whose data already sprawls across DealCloud, SharePoint and Exchange, and with a technical team able to build on the API, Metal is a credible answer to a real problem. Expect a 30 to 60 day project rather than a subscription you switch on, and expect to negotiate the contractual detail rather than read it.
- Choosing a selection results in a full page refresh.
- Opens in a new window.