01 · Graph-grounded
Answers grounded in 250K+ products and 1K+ signal sources.
APEX queries the live Product Graph · not the open web, not a static document index. Every claim links back to a typed node and a source URL.
APEX Copilot
Agentic AI, grounded.
Every answer cites the underlying node. Every citation traces to a source URL.
APEX is the copilot layer of the Product Graph · built on Graph RAG, not vector retrieval. It answers competitor, category, and customer questions with full provenance, so the answer survives an audit.
The agent swarm
One supervisor. 20 specialist agents. 156 graph tools.
APEX isn't one model · it's a coordinated swarm. The supervisor reads your intent and routes the right specialists. Watch a query fan out, light up the relevant agents, and assemble a cited answer.
You ask APEX
Build me a battle card for HubSpot vs. Salesforce in mid-market.
APEX returns
Waiting for specialist responses…
One supervisor coordinates 20 specialist agents, each with its own graph-grounded toolset (156 tools total). The supervisor picks the right specialists per query · you ask one question, the swarm responds.
What you can ask it
Eight kinds of insight, each with the receipts.
These are not demo prompts. Each is a question the engine answers today, with the mechanism named and the evidence chain attached — and half of them link to a page where you can run it yourself, no account needed.
Next-move prediction
“What will they do with the $200M they just raised?”
The two-sided engine builds a probability distribution over candidate moves from live signals, then reweights it by what their management is actually organised and funded to do (Porter Four-Corner priors). You get odds, not adjectives.
Runs live in your browser — drag the priors yourself
Run the engine →Blind-spot exposure
“What do they believe that their own customers contradict?”
APEX joins a competitor's stated strategy against what their buyers write in reviews — at the entity, not by keyword. A move premised on a belief the evidence contradicts scores as MORE exploitable, because they keep funding the mistake.
2.4M+ reviews, categorized and joined to 250K+ products
See the amplifier →Analyst-placement forecast
“Where do we land in the next Magic Quadrant?”
Placements from the editions you're licensed for are read into the graph — each one recording whether it was measured from the figure, computed from sub-scores, or estimated — then projected forward with the movement drivers named.
Six frameworks: MQ, Wave, MarketScape, PEAK, Critical Capabilities, Provider Lens
See the frameworks →Churn with named drivers
“Which accounts are at risk — and what exactly do we do?”
Eight lifecycle-agnostic dimensions score every account from public evidence alone, sharpened — never replaced — by your CRM. Risk is 0–100 where every point traces to a driver you can argue with and act on.
Eight dimensions per account, each with its evidence attached
See Customer 360 →Feature-gap economics
“Which gap is actually costing us deals?”
Win/loss fuses six sources — your closed deals, deal-bound interviews, surveys, support friction, public reviews, and the competitor's own surfaces — then ties each recurring loss reason to the capability gap behind it.
707,306 feature cells scored across 2.3M capability rows
See Win/Loss →What buyers actually say
“Is their onboarding really better than ours?”
Aspect-based sentiment models — run on our own infrastructure, not a third-party API — read every review at the feature level, so ‘ease of setup’ and ‘support quality’ are scored claims with the review text as the citation.
2.4M+ reviews scored at the feature level, 0–100
See the data foundation →Signal triage
“What happened this week that matters to us?”
1,000+ sources are watched on defined cadences. One article can split into several signals — a raise, a hiring plan, an upmarket push — each scored 0–100 for significance, entity-linked, and deduplicated across sources instead of stacked.
Every signal carries its source URL and lands on the entities it touches
See the signal engine →Market sizing & whitespace
“How big is this category, who's winning, and where's the gap?”
The graph already holds the market: every product typed against vendor, category, and capability. Sizing a category is a traversal, not a research project — and the whitespace is what the traversal doesn't find.
250K+ products across 2,546 categories, queryable free
Size a market now →Anatomy of an APEX answer
Question to cited answer in 5 stages.
Every APEX call follows the same lifecycle · whether you ask in Studio, in Slack, or from an MCP-connected client. Watch a real query travel from prompt to citation.
01 · Ask
User asks APEX
Which 3 CRM vendors are most threatened by AI-native entrants?
02 · Route
Supervisor agent routes intent
Picks 2 specialists: Competitive analyst + Innovation watcher.
03 · Retrieve
Multi-hop graph traversal
Product Categories (CRM) → Market Leaders → cross-ref with Funding Signals + AI Disruption score
04 · Cite
Citations assembled
8 source URLs · 3 review aggregates · 1 analyst quote · all returned with prompt + model + token hash
05 · Answer
Cited answer streamed
1. Salesforce · 2. HubSpot · 3. Microsoft Dynamics · with link to every claim.
The same lifecycle runs for every APEX call · in Studio, in Slack, and over MCP from any external client. Every citation traces back to a typed node on the graph; every prompt is hashed + auditable.
What APEX does
Four capabilities that keep agentic AI projects alive.
Gartner predicts over 40% of agentic AI projects will be cancelled by end of 2027 due to escalating costs, unclear business value, and inadequate risk controls (Gartner press release, June 25, 2025). APEX is the grounding layer those projects need to survive.
02 · Auditable
Every answer carries provenance and a citation trail.
Entity-level citations on every response. Hover any claim to see the underlying signal, source, and timestamp · ready for SOC 2 and EU AI Act audit logs.
03 · MCP-native
Connect your internal copilots through the MCP server.
APEX ships with a Model Context Protocol server, so your existing AI workflows (Claude, ChatGPT, internal agents) can ground answers in the same graph.
04 · Tenant-isolated
Row-level security on every read and write.
Your tenant's slice of the graph (battle cards, RFX responses, win/loss data) stays private. Public graph data improves through aggregate trends only.
APEX vs. a general-purpose LLM
Same question. Different answers. Different evidence.
A real prompt run against APEX (grounded in the Product Graph) and ChatGPT (general web knowledge). Note the citation count, recency, and specificity.
Prompt
“Who are the top 5 CSPM vendors with FedRAMP authorization and Salesforce integration, ranked by review count?”
APEX
Graph-grounded · 1.4s
Here are the 5 vendors meeting all three criteria, ranked by aggregated review count:
- 1. Wiz · 1,842 reviews [g2.com, ↗]
- 2. Palo Alto Prisma Cloud · 1,514 reviews [g2.com, ↗]
- 3. Lacework · 1,128 reviews [trustradius, ↗]
- 4. Orca Security · 904 reviews [g2.com, ↗]
- 5. Sysdig Secure · 612 reviews [g2.com, ↗]
Provenance
18 citations across 4 sources. All FedRAMP statuses verified against fedramp.gov (retrieved 2026-05-22). Salesforce integration confirmed via each vendor's integration directory.
ChatGPT
General LLM · 6.2s
Here are some leading CSPM vendors that may meet those criteria. Note that FedRAMP and integration details can change over time and I'd recommend verifying with the vendor:
- Wiz · widely regarded as a leader
- Palo Alto Prisma Cloud
- Lacework
- Orca Security
- CrowdStrike Falcon Cloud Security
I don't have specific review counts or up-to-date FedRAMP status for these. For ranking, you may want to consult G2 or Gartner directly.
Provenance
No citations. Training data cutoff ~2024-04. FedRAMP and integration status unverified.
18 → 0
Citations (APEX vs ChatGPT)
2026-05 → 2024-04
Data recency
4.4× faster
Time to grounded answer
Why Graph RAG
Vector RAG retrieves chunks. APEX traverses entities.
Most enterprise AI projects bolt a vector index onto a document corpus. That works for FAQs. It fails the moment a question requires joining two facts that live in different documents, or following a chain of relationships.
Vector RAG (what most teams ship)
Document chunks · cosine similarity · single hop
- Retrieval unit: Text chunk · no entity boundary
- Multi-hop reasoning: Weak · cosine similarity over chunks doesn't follow relationships
- Citation granularity: Document-level · “according to this PDF”
- Update cycle: Re-embed on every source change · expensive at scale
- Failure mode: Hallucinates the join · “trust me, these chunks are related”
APEX Graph RAG
Typed nodes · FK traversal · multi-hop with provenance
- Retrieval unit: Typed entity (Product, Vendor, Category, Job, Country)
- Multi-hop reasoning: Native · FK constraints traverse relationships across entity types
- Citation granularity: Cell-level · source URL + retrieval timestamp on each claim
- Update cycle: FK constraint propagates the change · no re-embedding required
- Failure mode: Returns “no path found” instead of hallucinating
The trade-off: Graph RAG requires the graph to exist. That's the foundation · PYRAMYD pre-built it across 90 universal node types, and the verified Product Graph. Vector RAG ships faster · the graph is what makes the answers survive an audit.
Ground your agents in the same graph our customers use.
Book a demo to see APEX answer your hardest competitor, category, and customer questions · with full provenance, in seconds.