Juicebox

Juicebox enumerates its integrations by name on the pricing comparison table: all 41 applicant tracking systems (Greenhouse, Lever, Ashby, Workday, iCIMS, SAP SuccessFactors, Jobvite, SmartRecruiters, Oracle Taleo, Cornerstone TalentLink, UKG Pro Recruiting, Workable, Teamtailor, BambooHR, Personio, Fountain, JazzHR, Recruitee and more) and all 21 CRMs (Salesforce, HubSpot, Microsoft Dynamics 365 Sales, Zoho, Pipedrive, Close, Zendesk, Affinity, Copper, Insightly, SugarCRM and more), alongside Gmail and Outlook for outreach: three ecosystem classes and 64 named systems.

The action is documented. The Agents page states Juicebox syncs with those systems without manual exports, agents create and send personalized outreach through the customer's own connected Gmail or Outlook mailboxes, and profiles export into ATS and CRM against a counted export limit. Those are authenticated writes into systems Juicebox does not own.

ATS integration, CRM integration and network sourcing are Business tier only. Free, Starter and Growth get CSV export and mailbox connection but no ATS or CRM sync, so this integration estate is unavailable on three of the four published plans.

Sourcejuicebox.ai/pricing and /agentsread 2026-09-12

The Agents page lays out multi agent orchestration step by step, under that heading.

A context intelligence layer pulls from the ATS, docs and intake calls; talent market mapping runs simulations across skills, titles and company stage to find the largest pool still matching the bar; search calibration builds the query through conversation; intelligent candidate analysis assesses every profile against custom criteria and company wide preferences and surfaces a shortlist; personalized outreach is written and sent; and real-time tracking follows response rates. Six named stages run end to end without a recruiter driving them.

Several agents deploy at once across different job openings, "each operating independently with tailored search, evaluation, and outreach logic", with ownership managed across every active agent in one dashboard. The sequence is customer configurable at named join points, with checkpoints at shortlist or sequencing. Participants are the agent, the recruiter owner who calibrates and approves, the hiring manager who reviews shortlisted leads on collaborative seats, and the candidate who receives outreach. The vendor's blog announces Agent 4.0 as its biggest release yet.

Sourcejuicebox.ai/agentsread 2026-09-12

Before it searches, the agent grounds itself in the customer's own material. In what it calls the context intelligence layer, before sourcing a single profile the agent "pulls context from your ATS, docs, intake calls, and past Juicebox searches" to understand the role the way the team does: four sources, three of them the customer's own material and the fourth the customer's own search history, read on every run. Past searches persist and projects are unlimited on every paid tier, so the store stands between runs and grows with use.

Alongside it sits the vendor corpus: 800 million profiles across 60+ data sources, queried in natural language with no Boolean, with smart profiles carrying linked data sources and AI inferred skills. Both persist, scale far past any context window and stay queryable.

Sourcejuicebox.ai/agents and /peoplegptread 2026-09-12

A configurable hold sits before the agent's two consequential actions. Flexible Autonomy Settings is a named product feature: "Run agents fully hands-off, or set manual checkpoints at shortlist or sequencing", with the line "you decide how much control you want." The pricing matrix corroborates it as a switch on the Agents add-on, "set to auto-email or auto-shortlist". A checkpoint at sequencing holds the agent before it sends outreach; a checkpoint at shortlist holds it before it advances a candidate.

Hands off is an available mode, so a customer can switch the gate off; the hold ships but is not mandatory.

Sourcejuicebox.ai/agents and /pricingread 2026-09-12

Certification and access controls are both documented, with sharp tier gating. The Agents page states "SOC 2 Type II certified" in its enterprise section, and a trust center at trust.juicebox.ai is linked from the footer alongside a status page, a responsible disclosure page and a privacy center with data access and deletion requests.

ISO 27001 and GDPR compliance appeared on juicebox.ai/trust as of July 2026. On access, SSO and access control authenticate users through the customer's existing identity provider, "assigning role-based permissions to control what each user can access", with SAML/SSO as its own line on the pricing matrix and admin seats carrying billing and activity visibility.

SAML/SSO and SOC 2 reporting are Business tier only, and the matrix notes that SOC 2 reporting must be requested from sales rather than downloaded. Free, Starter and Growth customers get neither the identity integration nor the report: the certification covers the platform, but the controls and the evidence do not reach three of the four published plans.

Sourcejuicebox.ai/agents, /pricing and trust.juicebox.airead 2026-09-12

A customer can inspect how the agent reached a recommendation, but no retained record of what it did is documented. Real-time performance tracking gives sourcing activity, response rates and agent ownership across every active agent in one dashboard, with usage analytics on the Business tier. The talent market mapping step lets a customer "understand how your agent got to its recommendation for each step", and AI Spotlight shows why a candidate matches a search, so the reasoning path behind a recommendation is inspectable stage by stage, not only the final score.

No audit trail, action log, run history or export is documented, and nothing states that what an agent did, which profiles it rejected, which emails it sent, when and under whose account, is captured and retrievable afterward. A live dashboard answers what is happening now; an audit trail answers what happened, and on a product whose agent can be set to auto-email unattended, that is the record a buyer would need.

Sourcejuicebox.ai/agents and /pricingread 2026-09-12

A preference set persists across runs. Agents assess every profile against custom criteria and "company-wide preferences", held at the organization level and applied by every agent, which is a customer authored store reapplied across runs and agents. A customer can recalibrate by chatting with the agent as they would a teammate, and "it adjusts without losing context", which is retained conversational state across a calibration; past Juicebox searches persist as an input to future ones. What is retained and where it applies are both named.

No lifetime or control is documented: no page states how long company-wide preferences or agent context persist, whether they can be inspected, edited as a set, exported or deleted independently of the account, or whether they are scoped per agent, per project or per organization beyond the phrase company-wide.

Sourcejuicebox.ai/agentsread 2026-09-12

No deployment or residency choice is offered on any commercial surface. The pricing comparison table itemizes the platform down to mailbox counts, export batch sizes and billing method, and carries no region, residency, data location or hosting row at any of the four tiers, not even on Business, where SAML/SSO, SOC 2 reporting and usage analytics are the listed enterprise differentiators.

Nothing on the agents page, the PeopleGPT page or the footer names a region or a customer controlled environment. The FAQ's only geography answer concerns the coverage of the candidate data (800 million profiles globally, with region specific sources being added), which is the product's reach rather than where the customer's own data sits.

A trust center at trust.juicebox.ai is linked from the footer alongside GDPR and CCPA; the product and pricing pages carry no residency or subprocessor commitment and no EU option.

Sourcejuicebox.ai/pricing, /agents and site footerread 2026-09-12

Reusable starting points are published, but not a set of distinct prebuilt agents. A Search Library at docs.juicebox.work/search-library, linked from the site footer under Resources, holds reusable searches a customer starts from rather than composing from scratch. AI email templates ship on every tier including Free, and multi agent flexibility lets a customer deploy several agents at once across different openings, "each with tailored search, evaluation, and outreach logic", so an agent is an instantiable, separately configured unit.

There is one agent type: the parallel agents are instances of the same sourcing agent pointed at different roles, not distinct prebuilt agents a buyer selects among. The Search Library's size, browsability and authorship are not documented.

Sourcejuicebox.ai/pricing, /agents and site footerread 2026-09-12

No operator needs to start the agents, but they reach candidates through one channel. Agents run 24/7 as the product's central claim, refreshing pipelines and sourcing continuously, and several agents run in parallel across different openings, each on its own schedule. The Agents add-on carries unlimited email credits so the cadence is not throttled, and follow-ups sequence automatically once outreach begins.

Outreach is email only. The pricing matrix lists Gmail and Outlook integration and nothing else, metered as one mailbox on Starter, three per seat on Growth and six per seat on Business. No SMS, WhatsApp, phone, voice, in platform chat, candidate portal or social channel is offered; phone numbers are sold as contact data at three credits each, data the recruiter receives rather than a channel the agent uses.

Sourcejuicebox.ai/pricing and /agentsread 2026-09-12

Juicebox offers no model choice and names no model provider for PeopleGPT, the Agents, or the AI Spotlight and AI filter features. Across the agents page, the full pricing comparison table and the site footer, no model picker, admin setting, provider list, routing description or bring your own key option is documented.

The pricing matrix enumerates the platform line by line, seats, searches, AI search builder, AI filters, talent insights, smart profiles, AI spotlight, contact credits, exports, integrations, mailboxes, SOC 2 reporting, SAML/SSO, usage analytics, billing and support, and carries no model or LLM row at any tier.

The AI Audit Center in the footer addresses how matching is audited for fairness, not which model runs it. Over the models running the agent, the customer chooses nothing and is told nothing.

Sourcejuicebox.ai/pricing, /agents and /peoplegptread 2026-09-12

The documentation is product help. Its surfaces, docs.juicebox.ai in the footer, a knowledge base at docs.juicebox.work and a help center at help.juicebox.work, along with a Chrome extension on the Chrome Web Store and 62 prebuilt ATS and CRM connectors, are not a build surface: every documentation link with a described destination points at product help, such as docs.juicebox.work/insights for Talent Insights, /usage-analytics for reporting and /search-library for saved searches.

No public API, endpoint list, authentication guide, SDK, webhook or MCP server appears on the product pages or in the footer, and the pricing comparison table, which enumerates integrations, exports, security and support line by line, has no developer row. The integration model is a fixed set of vendor maintained connectors plus CSV export, and nothing documents an outside system calling Juicebox.

Sourcejuicebox.ai/pricing, /agents and site footerread 2026-09-12

There is a tuning loop but no test harness. Agents assess every profile against the customer's custom criteria and company wide preferences, learn from recruiter approvals and rejections, and recalibrate through conversation as priorities shift, which is adjustment against observed output on the customer's own configuration. Usage analytics on the Business tier and a real-time dashboard of sourcing activity and response rates measure whether it is working.

The AI Audit Center in the site footer resolves to trust.warden-ai.com/juicebox, a third party audit of Juicebox's matching in general published as assurance; it tests the vendor's product rather than the customer's deployment. No test environment, sandbox, holdout, versioning or scored comparable result is documented, and there is no way to exercise a criteria change against a known set before the agent runs it on live candidates and emails them, which matters on a product whose agent sends outreach unattended when set to auto-email.

Sourcejuicebox.ai/pricing, /agents and the AI Audit Center link in the site footerread 2026-09-12

The first party Chrome extension, linked from the site footer to the Chrome Web Store, is a capture tool: the recruiter decides which profile page is open and the extension reads it into Juicebox. That is ingestion through a browser, not an agent operating one.

The agent itself queries an indexed corpus of 800 million profiles and sends email through the customer's connected Gmail or Outlook mailbox. Both are programmatic, neither navigates anything, and nothing documents a hosted or local browser, a desktop session, or remote or local computer control.

Sourcejuicebox.ai/agents, site footer and the Chrome Web Store listingread 2026-09-12