Unify

Named write targets plus Unify's own route to anything else.

The pricing table names Salesforce and HubSpot, read-only on Pro and read-write on Business, Slack notifications from Pro, managed Gmail and Outlook mailboxes and a dialer on Business, and "Open API + webhooks" with all integrations on Business; the developer documentation describes a Data API that manages objects and sends records, an Analytics API for event collection and a Bulk API, with Python and TypeScript SDKs.

Inward, more than 40 data vendors feed the signal layer with the provider shown per result. A documented route for the customer to connect further systems means the action surface does not stop at the two named CRMs, and gating it to Business limits availability rather than capability. The named write destinations remain few.

Sourceunifygtm.com/pricing and docs.unifygtm.com/developers/introductionread 2026-10-01

Plays are signal-triggered workflows that run end to end without a human step: when an account crosses an intent threshold the play fires, enriches the contact, adds them to a sequence and starts outreach, and the June 2026 agent release adds always-on agents running continuous custom signal detection across tens of thousands of accounts.

The plan table confirms automated plays as a shipped Business entitlement rather than a positioning claim, and agentic chat runs on every tier so the same agent can be driven conversationally or left to run. Around the chain sit account research, AI copywriting and personalization, multi-channel sequencing and a task dashboard, with managed mailboxes and deliverability handling the send. What fires is configured by a person in advance, so the autonomy is in execution rather than in deciding what to run.

Sourceunifygtm.com/product/plays, unifygtm.com/pricing and Unify next generation AI agents releaseread 2026-09-02

Agents research against a large aggregated data layer joined to the customer's own systems. The published figures are 1.1 billion people and 65 million companies drawn from more than 40 data sources, available on every tier, with Unify describing its approach as a vetted wholesale marketplace rather than a black-box data layer, stating that it evaluates every vendor before making it available, negotiates on the customer's behalf, and shows which underlying provider powered each result in the workflow.

That last point is unusual and is real grounding transparency: a user can see which source a fact came from. On top of the purchased layer sit the customer's own signals, CRM records from Pro upward, email activity, and first-party website intent and product usage on Business, and account research runs on every tier. Still, the corpus is largely Unify's purchased data about prospects rather than an ingested store of the customer's own positioning or collateral, so grounding is deep on who to contact and thin on what the customer sells.

Sourceunifygtm.com/product/b2b-company-contact-data and unifygtm.com/pricingread 2026-09-02

Oversight is exercised at design time and reinforced by a spend control, with nothing between the agent and the send. Humans author every Play, define the audience, map the signals and build the sequence, and setup is deliberately not plug and play, so the judgment sits in configuration rather than in review.

One real runtime guardrail is documented: Unify enforces a hard credit cap, so when seat credits and the account pool are both exhausted, credit-consuming actions pause rather than charge, with an in-product prompt to top up and any automation already in flight finishing its current step before stopping. That bounds the blast radius, with defined behavior at the boundary. The cap limits spend, though, not output.

No approval gate before an email sends, no confidence threshold, no review queue and no policy engine is documented, which is a notable gap for a system whose plays fire and send on their own once a signal crosses a threshold.

Sourceunifygtm.com/pricing and unifygtm.com/product/playsread 2026-09-02

Customer controls and an attestation stated plainly with its type are both published. Unify's pricing FAQ states in its own words that it maintains SOC 2 Type II compliance, performs real-time threat monitoring and regularly contracts external security audits, with a security contact address published, and a public trust portal at trust.unifygtm.com carries the same.

Controls: SSO is documented, listed as an add-on on the Business tier, with SAML 2.0 through Okta or Azure AD, and team management appears from Pro. The legal set is complete and linked from every page: terms, privacy policy, a customer data processing agreement and the trust portal. Not found: any ISO certification, encryption detail, sub-processor list or report request process. SSO being an add-on rather than included limits availability rather than capability.

Sourceunifygtm.com/pricing, trust.unifygtm.com and unifygtm.com/legalread 2026-09-02

Reporting is tiered and real; agent auditability is absent. The published entitlement table names sequence analytics and rep activity from Pro, and signal analytics and pipeline attribution on Business, alongside a task dashboard on every tier and deliverability analytics on Business.

Agent runs are metered individually in credits at one credit per run, which gives a per-run cost record and therefore a countable history of how often agents executed, and Unify documents what happens at the boundary, with in-flight automations finishing their current step when credits are exhausted. What is missing is anything agent-facing: no decision trace, no reasoning log, no record of which signal or source drove a given play to fire, no configuration audit trail and no retention policy. A buyer can see what the campaigns achieved and how many runs were consumed, and cannot reconstruct why a particular account was worked.

Sourceunifygtm.com/product/analytics and unifygtm.com/pricingread 2026-09-02

A persistent data layer, not agent memory. Account, contact and signal history sit in Unify's data platform and agents read them on every run, so a play acting on an account today sees what was recorded about it before; credits carry over for up to twelve months and top-ups pool across seats. Those are business records and commercial state rather than memory, and the data layer serves as the agents' knowledge source. No memory construct the agents write and read back, no learned preferences and no retention or forgetting control at the agent level is documented.

Sourcedocs.unifygtm.com/developers/introduction and unifygtm.com/pricingread 2026-10-01

Delivered purely as multi-tenant SaaS with no deployment or residency control.

No region selection, EU option, on-premises route, single-tenant tier or private VPC appears on the pricing page's full entitlement table, which lists plan differences down to mailbox counts and integration read-write scope, nor on the product pages, nor on the public trust portal where a residency commitment would normally sit alongside the SOC 2 statement.

These are surfaces that had the opportunity to say otherwise. One place is still unread: the customer data processing agreement linked in the footer could hold a processing-location or transfer statement.

Sourceunifygtm.com/pricing, trust.unifygtm.com and unifygtm.com/legal/customer-data-processing-agreementread 2026-09-02

Plays give a reusable workflow structure and named agent capabilities ship for account research and copywriting, so a customer is not starting from an empty canvas.

But agents are prompt-authored by the customer rather than drawn from a catalog, and nothing is browsable: no play library, no template gallery, no marketplace and no importable or clonable configuration appears on the product pages or in the published entitlement table, where automated plays is a single Business line rather than a set to choose from.

Customers get the means to build agents but nothing to browse and select. Configuration is meaningful work rather than a one-click start, which matches Unify's own positioning.

Sourceunifygtm.com/product/plays and unifygtm.com/pricingread 2026-09-02

Triggering is the product. The plan table enumerates the signal classes by name: tech stack, new hires and open roles, social signals, lookalike signals, warm intro paths, news and company events, fundraising and growth stage, paid ad activity, SEO and AI search visibility, brand and sponsorship activity, e-commerce data, and SMB and local business data on every tier, with email activity and CRM signals from Pro, and first-party website intent and product usage signals on Business.

Those feed signal-triggered automations on Business, which fire at configurable intent thresholds, and the June 2026 agent release adds always-on detection of custom buying signals across tens of thousands of accounts, which wakes the agents on their own. On the outbound side, multi-channel sequences and social messaging ship on every tier, a dialer is in beta on Business, and Business adds managed Gmail and Outlook mailboxes.

Sourceunifygtm.com/pricing and the Unify AI agents releaseread 2026-10-01

The provider is named and the top tier carries more than one model. The plan comparison table has a row headed "LLM model access": GPT 5.4 on Free, Base and Pro, and GPT 5.4 and 5.6 on Business, with the tier bullets reading standard models on the lower plans and advanced models on Business.

So a buyer can see which model runs their work and can get access to a second by moving tier, and Unify states plainly that AI usage is billed through the credit balance because it passes through the cost of running the foundation models, with no separate AI fee. Naming the model and version, with access to two of them, is a real if coarse form of choice. The choice is made by choosing a plan rather than a model.

No per-run or per-agent selector, no routing policy, no bring-your-own-model and no bring-your-own-key is documented, and the table does not say whether a Business customer picks between 5.4 and 5.6 or whether the platform decides.

The June 2026 agent release describes Unify benchmarking open-source and frontier models across thousands of real customer workflows before selecting one to power its agents, which is Unify choosing and sits alongside the tiered access above.

Sourceunifygtm.com/pricing and Unify next generation AI agents releaseread 2026-09-02

A full developer surface, including a documented MCP server. Unify runs a remote MCP server at mcp.unifygtm.com/mcp with OAuth, documented at docs.unifygtm.com with setup paths for Claude Desktop and claude.ai including the workspace-admin connector flow, and for Claude Code, Cursor and Codex through an official plugin at github.com/unifygtm/agent-plugins that bundles Unify skills alongside the MCP configuration.

Unify states that a connected assistant can research companies, build lists, write sequences and update the CRM using the same agent that powers Unify Chat, so the interface carries the product's real capability rather than a read-only view.

Around it: a public developer documentation site with an llms.txt index, Platform APIs and Intent APIs, an official TypeScript library and a public GitHub organization publishing client libraries including a React binding for the intent client, and Open API plus webhooks listed as a Business-tier entitlement.

Two limits, both about availability rather than absence: MCP is in early access and under active development, and of the documented APIs only the Data API is generally available, with the others in beta.

Sourcedocs.unifygtm.com developer documentation and MCP guide, github.com/unifygtmread 2026-09-02

The evaluation discipline is real and points inward. Unify documents building offline benchmarks and online evaluations, testing open-source and frontier models across architectures, and running large-scale evaluations across thousands of real customer workflows before selecting the model that powers its agents. That is rigorous and it is Unify testing Unify.

What the customer gets is measurement of outcomes rather than of agents: sequence analytics and rep activity from Pro, signal analytics and pipeline attribution on Business, plus a 14-day free trial of Pro that works as try-before-you-buy rather than a test harness. No sandbox, dry run, simulation, output scoring, regression check when a play or prompt changes, or comparison between agent configurations is documented anywhere. Customers can measure the results of an agent's work, but they have no evaluation surface of their own.

Sourceunifygtm.com/pricing and Unify next generation AI agents releaseread 2026-09-02

No browser control or computer use is documented. Two capabilities needed checking and neither qualifies. Website visitor identification is passive tracking through an installed intent client running on the customer's own site, which is telemetry collection rather than an agent operating software.

A browser extension ships on every tier, but it brings Unify's data and actions to the user inside pages they are already viewing rather than driving an interface unattended. Everything else runs through APIs: enrichment, sequencing, CRM sync and the managed mailbox layer.

No headless session, screen operation or robotic process automation appears on the product pages, the entitlement table or the developer documentation.

Sourceunifygtm.com/pricing and unifygtm.com/products/signalsread 2026-09-02