UserGems

Authenticated action across genuinely different classes of system, not depth inside one. Outward the agents create and enrich records in Salesforce and HubSpot, enrol contacts into sequences in Outreach and Salesloft, sync contact-level audiences to LinkedIn, Google and Meta, post alerts to Slack, and trigger physical gifts through Sendoso.

Inward they ingest 21+ native signal types plus the customer's own first-party material including call transcripts, product usage, events and content engagement, and pull call transcripts from Gong and matching customer stories from a Webflow CMS mid-run. Crucially there is a documented route to add more: any MCP-compatible tool can be wired into a workflow and called by the agents during a campaign, which is an open extension path rather than a fixed connector list. That spans CRM, sales engagement, advertising, messaging, conversation intelligence, content management and gifting.

Sourceusergems.com/product/mcp and usergems.com/product/overviewread 2026-09-02

Specialized agents running a named chain end to end, on a schedule, across systems the vendor does not own. The vendor's own worked campaign detects past-champion movements at ICP accounts, enriches or creates the records in Salesforce, writes a personalized five-email sequence, enrols the contacts into a named Outreach sequence, syncs them to a LinkedIn ad audience and triggers a gift through Sendoso for anyone scoring above 85, then repeats every Monday without being asked again.

A second runs on closed-lost accounts that reach Tier A, pulling Gong transcripts and the right Webflow case study before drafting. Output passes between agents rather than between separate tools: data agents capture and enrich, intelligence agents score and prioritize, the writing agent drafts from that context and the orchestration agent coordinates outbound, ads, alerts and CRM writes. Set once, runs forever is the vendor's own framing and it is documented as shipped behavior with named steps.

Sourceusergems.com/product/mcp and usergems.com/product/overviewread 2026-09-02

A maintained structure built out of the customer's own history, read by the agents on every run. UserGems analyses the customer's opportunities, customers and deal outcomes to derive 600+ ICP criteria and to identify personas and value proposition, and that model is continuously updated as outcomes accumulate rather than assembled per request. The customer's own material is ingested directly: call transcripts, email exchanges, product usage, event attendance and content engagement.

The research agent reads that corpus rather than relying on what reps wrote down, analyzing call transcripts and email exchanges to recover a buyer's goals, objections and pain points and carrying them into the messaging, and the writing agent is additionally trained on uploaded brand guidelines, best-performing emails and customer stories. Grounding is therefore on the customer's own knowledge on both sides, who they sell to and how they sell.

Sourceusergems.com/product/overview and usergems.com/product/ai-emailsread 2026-09-02

Real control at design time, and nothing standing between the agent and the send. The customer can edit the scoring weights directly, so the targeting decisions that determine who gets worked are set by the team rather than by the model, and Gem-E is trained on uploaded brand guidelines, best-performing emails and approved customer stories, which bounds what it may produce. UserGems ships no send capability.

Drafts are enrolled into Outreach or Salesloft, so the review-and-send step belongs to the sequencer the customer already runs, not to UserGems. No approval step, review queue, hold or confidence threshold is documented inside UserGems, and the always-on campaigns run the other way entirely, detecting movements, creating CRM records, writing outreach, enrolling contacts and triggering gifts on a weekly schedule with no human step in the loop.

Sourceusergems.com/product/mcp and usergems.com/product/ai-emailsread 2026-09-02

The customer controls who gets in, and the compliance documentation sits behind it.

The security page names SAML single sign-on through the customer's own identity provider, stating explicitly that an organization controls authentication to UserGems and enforces its own password policy, account recovery and multi-factor requirements, alongside Sign-in with Outreach via OpenID, least-privilege access to customer data and MFA on administrative access.

Behind that sit AES-256 encryption at rest, TLS and HSTS in transit, KMS-protected secrets, annual third-party penetration testing, a Salesforce AppExchange security review including a further independent test, and a SafeBase trust portal publishing a SOC 2 Type II report alongside a pentest report, subprocessor list, data processing agreement, SIG Core and CAIQ questionnaires, data flow diagram and certificate of insurance, with GDPR, CCPA and an EU AI Act limited-risk classification.

The security page states active SOC 2 without a type; the Type II is stated on the trust portal. Vendor of record is ShelfFlip, Inc. trading as UserGems; the attestation and the trading name sit with the same entity.

Sourceusergems.com/legal-security/security and trust.usergems.comread 2026-09-02

A real why, covering one decision and no others. Transparent scoring is a shipped capability rather than a claim: accounts are mapped into buying stages with customizable weights and automatic decay, and teams can see exactly why an account sits in a given stage, which signals moved it there and how close it is to sales-ready, positioned against opaque intent-only scoring. Alongside it, reporting covers TAM progression, pipeline generated and return on dollar spent. Prioritization is one decision among many, and nothing traces the agent's actions.

No record shows what Gem-E wrote for a given contact and on what basis, what the research agent read, which campaign step fired, or what was enrolled and when, and no log surface, export, retention setting or configuration audit trail is documented. A customer can reconstruct why an account was chosen and not why it was approached the way it was. The security page documents central logging of API and operating-system calls fed through detection rules; that is the vendor's own internal monitoring rather than a customer-facing audit trail.

Sourceusergems.com/product/overview and usergems.com/ai-agents/gem-e-for-abmread 2026-09-02

State the agent writes and reads back, bounded to the run it is in. When Gem-E composes a multi-step sequence it knows which signals it has already used and refers to its own previous emails so the thread reads as continuous, and it remembers previous steps and adjusts later messaging based on the reply received or on signals that arrived since. That is the agent consulting a record of its own prior output. Around it, always-on campaigns persist their configuration and re-run on schedule indefinitely once set.

The persistent account, contact and signal store synchronized with the CRM is a shared data store rather than a memory layer. What persists is within-sequence and campaign-level. No cross-session recall, no per-agent scratchpad carried between campaigns, no learned preferences accumulating across runs, and no retention, inspection or forgetting control exposed to the customer is documented.

Sourceusergems.com/product/ai-emails and usergems.com/product/mcpread 2026-09-02

Multi-tenant cloud in one fixed location, with the vendor stating so plainly. The security page says UserGems leverages data centers of a cloud service provider in the United States, and adds that customer data is not authorized to leave the production service environment. The trust portal's own summary is AWS US hosting with EU data transfer documentation, which is legal cover for moving data out of Europe rather than an option to keep it there.

No region selection, EU or other non-US option, private VPC, single-tenant tier, on-premises route or install inside the customer's own tenant is published. Knowing where the single region is tells a buyer where the data sits; it does not give them a choice. The VPC and firewall configuration described on the security page is UserGems' own production network, not a customer-controlled deployment.

Sourceusergems.com/legal-security/security and trust.usergems.comread 2026-09-02

Named agents arrive built and there is a great deal to read, but nothing to adopt. The platform ships a fixed roster the customer does not author: Signals, Research, Account and Contact Discovery and Data Hygiene as data agents, and Scoring, List Building, Personalization, Orchestration and Advertising as intelligence agents, on top of 21+ native signal types.

The Playbooks library is browsable and filterable by team, with each play tagged to the signal it runs on, but it sits under Resources beside the blog, guides and webinars, its entries are authored articles carrying bylines and named-person collaborations, and a customer reads one and then builds the campaign themselves rather than selecting a packaged configuration into the product.

The MCP page publishes three worked setup prompts a customer can paste to stand up an always-on campaign, which is closer to an adoptable asset than the playbooks are, but three examples on a product page are not a catalog.

Sourceusergems.com/playbooks and usergems.com/product/overviewread 2026-09-02

Work arrives from several genuinely different directions and leaves by several more.

The agents take in 21+ native signal types including job changes and champion movements, buyer intent, website visits, role changes, hiring, funding, M&A and 10-K filings, plus contact-level resolution that identifies which individual is researching rather than only that an account is in market, plus the customer's own first-party feeds of marketing engagement, form submissions, event attendance, product usage and NPS, and a research agent for custom signals.

Alongside signals, campaigns run on a clock, with the vendor's worked examples firing weekly and monthly. Results go out as email sequences through connected engagement platforms, contact-level ad audiences to LinkedIn, Google and Meta, Slack alerts, CRM record creation, and insights surfaced in the rep's browser during calls.

Sourceusergems.com/product/overview and usergems.com/product/mcpread 2026-09-02

The customer does not choose the model, and the vendor does not say which one it uses. No model selector, per-agent choice, routing policy, bring-your-own-model or bring-your-own-key is published, and no provider is named for the models powering Gem-E. The MCP page names Claude and ChatGPT and mentions GPT-4o, but those are the assistants the customer prompts from, calling UserGems as a tool. That is the customer picking the interface that reads the vendor, not the customer choosing what powers the product.

Sourceusergems.com/product/mcp and usergems.com/product/overviewread 2026-09-02

Outside callers can drive the platform, and the vendor ships that as a product. UserGems MCP is added as an MCP server in any MCP-compatible interface, with Claude and ChatGPT named, and the vendor draws the distinction that matters to a buyer: it executes rather than returns data, so a connected assistant builds campaigns, creates ad audiences, enrols contacts into sequences, drafts outreach and updates the CRM using the same agent that runs inside the product.

It also accepts inbound connections from the customer's existing GTM tools and calls other MCP servers mid-workflow, and the vendor documents the build case explicitly, feeding a customer's homegrown model the signals, scored accounts and enriched contacts it needs. A product-level API page exists on the property as its own offering, and the vendor announced the MCP in its own news index. A help center runs at help.usergems.com, but no public developer documentation site, SDK, client library or endpoint reference is published, and the MCP is documented on a marketing product page rather than in developer docs.

Sourceusergems.com/product/mcp and usergems.com/product/apiread 2026-09-02

A dashboard, not an engine. Reporting covers TAM progression, pipeline generated and return on dollar spent, and reply rates are measured against the agent's own output, but reporting is not a mechanism however detailed it is. Nothing runs a controlled comparison against a criterion: no A/B test of message or sequence variants, no holdout, no scored test set, no sandbox, dry run or regression check when a prompt or scoring model changes.

The nearest candidate is the writing agent adjusting messaging in light of a reply or a newly arrived signal, which is adaptation inside a single sequence rather than an experiment, and it is the vendor's model doing it with no surface the customer drives. A customer can therefore read what happened and cannot test anything before it reaches a prospect. The money-back ROI guarantee is a commercial warranty, and the vendor's own claim of 6-20% reply rates against an industry average is marketing rather than a result the customer can reproduce.

Sourceusergems.com/product/overview and usergems.com/product/ai-emailsread 2026-09-02

The one candidate is an extension that assists rather than operates. The vendor describes the AI Chrome extension as bringing AI assistance into the reps' browsers so they get insights while working in their CRM, sales engagement tools or during calls, with the stated purpose that there are no new tools for reps to learn. It surfaces signals and context into a page a person is already looking at and the person acts; nothing describes the agent driving an interface itself.

Everything else the platform does runs through programmatic routes: CRM writes, sequence enrollment, ad audience sync, Slack alerts and gifting all go through integrations, and the MCP executes against the product's own surface. No headless browsing, virtual desktop, screen or click automation, or navigation of a third-party console appears anywhere on the property.

Sourceusergems.com/product/overviewread 2026-09-02