Greenhouse

Greenhouse's partner estate takes action in systems Greenhouse does not own. It runs its own integrations marketplace at integrations.greenhouse.com, listing 520 integrations in four named classes, from HRIS and onboarding and job distribution to productivity and collaboration and candidate experience, with Alliance Partner and Preferred Partner tiers and localization across more than thirty languages.

The action is outbound. Job distribution publishes job posts to external boards, and HRIS and onboarding integrations push hires into the customer's HR system. E-signature, scheduled take-home tests and CLEAR ID verification each run in a third party service and return a result into the record, and SMTP relay with self service custom SMTP sends through the customer's own mail infrastructure. Multi brand job boards and featured internal job posts extend the same publishing path.

The count of 520 and the four category names are the marketplace's own, shown in its filter panel.

Sourceintegrations.greenhouse.com and greenhouse.com/pricingread 2026-09-08

A workflow engine the customer configures runs hiring across multiple participants. Structured hiring stages come with configurable interview kits and stage transitions, scheduling automation weighs interviewer limits, preferences, availability and working hours, and job and offer approval chains, bulk offers and candidate self scheduling round it out. Recruiter, coordinator and interviewer, hiring manager, approver and candidate are distinct participants, and work passes between them on defined stages. Auto-reject and auto-advance add unattended branching on Plus and above.

The agentic layer that would run this end to end has not shipped. Greenhouse's own features page says the Job Setup Assistant and the Candidate Insights Assistant are expected to launch in Q3 2026, with Analytics and Chart Builder in open beta and AI Report Insights expected in August 2026. What ships is assistive, from notes and charts to summaries and matching, on top of the configured workflow engine rather than an agent that sequences the work. The page still reads in the future tense after the August date has passed, which points to a stale page rather than a launch.

Sourcegreenhouse.com/greenhouse-latest-features and /pricingread 2026-09-08

The customer's whole hiring estate lives in Greenhouse as the system of record: candidates, prospects and prospect pools; applications, jobs, interview kits and scorecards; and interviewer feedback, offers, activity history and EEOC and demographic data. That corpus persists between runs, scales across an entire hiring organization far past any context window, and stays queryable in natural language.

Greenhouse MCP lets approved assistants query it directly to generate weekly health briefs, executive summaries and pattern monitoring, and the Analytics Chart Agent turns plain text questions into visualizations over the same reporting corpus. Notetaker writes back into it by mapping AI generated notes onto scorecard questions. The announced Candidate Insights Agent, designed to give answers linked to sources, is expected in Q3 2026 and has not shipped.

Sourcegreenhouse.com/greenhouse-latest-features and /pricingread 2026-09-08

The hiring decision stays with the customer, and approval chains ship, but nothing gates what the AI does. One of five published AI design pillars says Greenhouse AI can inform, summarize and surface insights, but the decision is always yours, not handed to an algorithm, and the FAQ repeats that AI does not make hiring decisions and people remain responsible for evaluating candidates. Voice AI follows the same shape: the customer sets the criteria, decides who moves forward, and can calibrate all scoring. Job and offer approvals ship on all three tiers.

Those approval chains gate a person's business action, posting a job or extending an offer, not an agent's action. Nothing documents an approval queue, confidence threshold or hold before any Greenhouse AI capability acts, and auto-reject and auto-advance rules run with no gate at all. Framing such as AI built to strengthen structured hiring and keep humans in control is posture rather than a mechanism.

Sourcegreenhouse.com/greenhouse-latest-features and /pricingread 2026-09-08

Attestations and access controls are both documented. Annual third party audits cover SOC 1 Type 2 and SOC 2 Type 2, Greenhouse is certified for ISO 27001:2022, ISO 27701:2019 and ISO 42001:2023, and it cites GDPR, CCPA/CPRA and PCI DSS along with the EU-U.S. DPF and its UK and Swiss counterparts.

The program around them includes annual third party penetration testing and a public bug bounty on HackerOne, security in the development cycle and encryption in transit and at rest, and a dedicated security team with a 24/7/365 security operations center. A trust portal is published at trust.greenhouse.com.

The access controls are named as product security features and confirmed on the pricing matrix: single sign on on all three tiers, SCIM user provisioning and granular user permissions, plus permissions approvals, interviewer groups and an audit log API. Tiering matters to a buyer: SSO and granular permissions are on Core, automated user management on Plus, and the audit log, configurable session time-outs and restricted file settings are Pro only.

Sourcegreenhouse.com/security and /pricingread 2026-09-08

Platform auditability and an explainability commitment are documented, but no trace of an agent run. An audit log is a named product security feature, exposed as an audit log API on the Pro tier, giving a retrievable record of activity in the system. One of five AI design pillars commits that every Greenhouse AI output is transparent and grounded in observable signals, so a team can audit, learn and hire with confidence, with structure described as replacing black box risk with something explainable and consistent.

Neither records what an agent did. The audit log covers user and system activity, and nothing states that AI or MCP actions are captured with their inputs, prompts or reasoning; explaining an output is not reconstructing what an agent did.

On a platform where outside assistants can drive the system of record through MCP, the record of what those assistants changed is what a buyer would ask for, and no page confirms it is captured. The audit log is Pro only, so most customers do not have it. Reports, dashboards, AI report filters and the BI Connector measure hiring operations, not agent behavior.

Sourcegreenhouse.com/security, /pricing and /greenhouse-latest-featuresread 2026-09-08

No agent memory layer is documented. The features, security and API pages and the pricing feature matrix show no state store across sessions, retained preference set, or memory scope and lifetime.

Candidates and applications, scorecards and notes, offers and activity history persist in Greenhouse, but that is the application's own data model: deleting it deletes the hiring record itself, which is what separates a data model from a memory layer.

The documented AI capabilities are stateless against it: Notetaker transcribes an interview, the Chart Agent answers a question and MCP serves a query, and nothing says any of them retains anything between invocations.

The Pro tier's data configuration controls cover session time-outs, restricted files and application limits, with no retention or deletion control over agent memory.

Sourcegreenhouse.com/greenhouse-latest-features, /security and /pricingread 2026-09-08

A single cloud footprint is disclosed, with no choice on top of it. The security page says network infrastructure and computing are provided by Amazon Web Services in a multi tier network environment, with routine backups, disaster recovery testing and uptime SLAs. The security page, the pricing feature matrix and the API page show no region list or regional hosting option, no VPC, on premises or customer tenant, and no way to select residency.

Cross border data transfers are stated to be subject to the EU-U.S. Data Privacy Framework, the UK Extension and the Swiss-U.S. DPF. Those are legal transfer mechanisms, not residency: a DPF certification tells a buyer nothing about where their data sits and gives them no choice about it. A residency commitment, if one exists, would most likely sit in the trust portal at trust.greenhouse.com.

Sourcegreenhouse.com/securityread 2026-09-08

Several separately named AI capabilities ship. Greenhouse Notetaker, on Core, Plus and Pro, records and transcribes interviews and maps AI notes onto scorecard questions, and Greenhouse Voice AI is a two way voice interviewing product with its own page and primary navigation entry. Greenhouse MCP serves Site Admins, the Analytics Chart Agent is in open beta, and Real Talent adds talent matching, fraud and spam detection and CLEAR ID verification.

They arrive bundled by pricing tier rather than as a set a buyer selects among. Notetaker, matching and MCP come with Core, while fraud detection, CLEAR ID and sourcing automation come with Plus; nothing is adopted, declined or swapped independently, and Core, Plus and Pro are a pricing structure rather than a pack.

There is no browsable catalog of agents, template library or surface to clone and configure one; integrations.greenhouse.com is a marketplace of third party connectors, not a library of Greenhouse agents. Two of six announced capabilities, the Job Setup Assistant and the Candidate Insights Assistant, are still expected in Q3 2026 on the vendor's own page.

Sourcegreenhouse.com/pricing and /greenhouse-latest-featuresread 2026-09-08

Event and scheduled triggers live in the workflow layer. Auto-reject and auto-advance rules act on an application without a person touching it, stage transitions fire on pipeline movement, and scheduling automation rules run off interview state. Scheduled take-home tests and candidate interview reminders fire on time rather than on a click.

Channels run from email automation with SMTP relay and texting to a candidate portal at MyGreenhouse with self scheduling and scheduling links, plus internal and multi brand job boards, e-signature and configurable candidate surveys, and Greenhouse Voice AI adds a voice channel for two way structured screening interviews at the top of the funnel.

None of the AI capabilities is triggered by an event. Notetaker, the Analytics Chart Agent and MCP are all invoked by a user or a connected outside assistant, so the rules that fire on events are classic ATS workflow automation and the AI sits behind a person's hand. Auto-reject, auto-advance, email automation and texting are on Plus and above.

Sourcegreenhouse.com/pricing and /greenhouse-latest-featuresread 2026-09-08

Nothing is documented about the models behind Greenhouse's own AI. The features, security and API pages and the pricing matrix show no model picker or admin setting, no provider list or routing description, and no way to bring your own model, and no model provider is named for Notetaker, Voice AI, the Analytics Chart Agent or Real Talent matching. The published AI principles speak to structure, explainability and decision ownership rather than model composition.

Greenhouse MCP lets a customer choose which approved assistant, such as Claude, Gemini or Copilot, connects to Greenhouse. That is a choice of assistant at the access layer, an interface for the customer's assistant to call Greenhouse, not a choice of model inside Greenhouse's own AI. The AI principles and ethical principles pages may name a provider.

Sourcegreenhouse.com/greenhouse-latest-features and /pricingread 2026-09-08

A full developer estate is published. The Harvest API has reference documentation at harvestdocs.greenhouse.io, and the Job Board API, the Onboarding API and further surfaces are indexed at developers.greenhouse.io. The APIs page states their purpose in the vendor's own words, to let customers and partners access and update data from Greenhouse, and names the write verbs: update candidate information, add attachments to candidate profiles, and advance, move and reject applications.

Around the APIs sit a Business Intelligence Connector and job custom fields available only through the API on Plus, a developer sandbox with sandbox sync on Pro, and a Partner Resource Center for partner integrations. Greenhouse MCP, available to Site Admins on Core, Plus and Pro, connects approved tools such as Claude, Gemini or Copilot to Greenhouse data in a structured, permission aware way. On both surfaces the call goes inward, into Greenhouse, and the documented APIs go well beyond the MCP server.

Sourcegreenhouse.com/api, harvestdocs.greenhouse.io, developers.greenhouse.io and greenhouse.com/pricingread 2026-09-08

A sandbox and a tuning loop are documented. A developer sandbox with sandbox sync ships on the Pro tier, so a customer can exercise configuration and integration changes before they reach production, though a sandbox tests whether a change installed correctly rather than what an agent did.

In the tuning loop, Greenhouse says all Voice AI scoring can be calibrated and will improve over time, that the customer sets the criteria for what makes a candidate great, and that on Talent Matching recruiters choose which criteria to prioritize and the relative weight of each factor. That is adjustment against observed output, with no documented holdout, scored test case or comparable result.

Greenhouse also publishes a bias audit statement covering Talent Matching, with disparate impact analysis and demographic variable testing by Warden AI across ten protected classes using the auditor's own proprietary dataset, published to a single public Greenhouse dashboard.

That tests the product in general, not the customer's deployed instance on the customer's data, and alignment to NYC LL-144, Colorado SB205 and California FEHA is a legal mandate, not evidence of whose artifact was tested. Reports, dashboards, AI report filters and the BI Connector measure hiring operations, not agent behavior.

Sourcegreenhouse.com/pricing, /bias-audit-statement and /greenhouse-latest-featuresread 2026-09-08

No browser or computer use is documented. The Prospecting Chrome plugin ships on all three tiers and sits in the Sourcing and CRM section of the pricing matrix alongside prospect creation and prospect pools, but with it the recruiter finds the profile and opens the page, and the plugin captures what is on it into Greenhouse. That is ingestion through a browser, not the product navigating, clicking or filling on its own.

The product running in a browser is not an agent operating a browser. No hosted or local browser, desktop session, or remote or local computer control appears on any of Greenhouse's product, security or developer pages.

Sourcegreenhouse.com/pricingread 2026-09-08