Fin vs Lorikeet (2026)

Fin runs on the customer's existing helpdesk across Intercom, Salesforce, HubSpot, Freshworks, Dixa, Front, Zoho, Sprinklr and Gorgias, adds Data Connectors for live external data, takes action to update external systems, and spans messaging and telephony channels.

Sourcefin.ai/pricingread 2026-09-05

Built in connectors cover Zendesk, Zendesk Messaging, Intercom, Salesforce (including Salesforce Messaging), HubSpot, Front, Help Scout, Kustomer, Slack and Genesys Cloud, and any helpdesk that sends webhooks connects through Flexible Ticketing. Telephony connects through Amazon Connect, Twilio, Vonage, Five9, AirCall, Dialpad and Talkdesk, and knowledge through Confluence, Guru, Google Docs and Notion.

Business systems connect through APIs, webhooks and MCP servers (Notion and Linear are named), so the agent reads data and takes actions such as processing a refund in Stripe or updating an account, with read only or write access set per integration, tools scoped to least privilege, short lived single operation tokens and HMAC signed inbound webhooks. Coach can set up custom API integrations, while API keys and secrets are entered by the customer.

Sourcelorikeetcx.ai/integrationsread 2026-11-02

Procedures are configured multi-step workflows the agent runs end to end, including branches that end in a handoff to a human or into a workflow. Workflow automations on the Intercom helpdesk and Operator, a separate agent for customer operations work, sit alongside.

Sourcefin.ai/pricingread 2026-09-05

The Concierge agent runs multi step work by following the customer's own standard operating procedures, whether a three step form or a forty step refund flow, inside natural language and deterministic workflows. Sensitive steps such as payments and identity checks run as code the agent can invoke but never alter.

Workflow steps include workflow matching, ticket triage, gather, get status, choice and route based on problem, with tool calls such as retrieving recent transactions shown inline, and a team of specialist agents can coordinate with third parties by phone, SMS, email and Slack to resolve multi party issues.

Customers deploy topic by topic, so the first topics go live while harder ones are still being trained, and Dreamweaver works from a goal such as retention rather than a written workflow.

Sourcelorikeetcx.airead 2026-11-02

Channels include Voice, Email, Messenger, Slack, WhatsApp, SMS, Instagram, Facebook Messenger and Fin over API. Tickets, cases and emails come in from the connected helpdesk, and a proactive outbound suite runs on the Intercom helpdesk.

Sourcefin.ai/pricingread 2026-09-05

Chat, email, voice, SMS and WhatsApp run on one agent, inbound and outbound, with an SDK and the Lorikeet chat widget as further surfaces; WhatsApp is served through the customer's Twilio number registered for the WhatsApp Business API. Voice answers and places calls on local numbers in the US, UK and Australia with sub second responses, and every channel inherits the same workflows, guardrails and quality scoring.

The outbound agent starts conversations itself: reminders, document requests, appointment confirmations and follow ups over SMS and voice, on recorded consent per channel, capped at three campaign calls per number in 24 hours, with voicemail detected and handled. Dreamweaver reaches out on a schedule or when something changes, on the channel each customer prefers, or chooses not to contact them.

Sourcelorikeetcx.airead 2026-11-02

Fin is trained on Knowledge Sources, Data Connectors, Procedures, Guidance and Categories, ingesting help center content from Intercom, Zendesk or Salesforce plus any public URL. The ingested content re-syncs every 24 hours for connected help centers and weekly for external web content.

Sourcefin.ai/help/en/articles/13975777-pricing-and-usage-limitsread 2026-09-05

The agent is grounded in the customer's own documented procedures, not retrieval alone. It learns from the customer's help docs, SOPs, policies and brand guidelines, taps into its systems for live context, and follows the procedures themselves rather than only searching articles.

A built in grounding guardrail, always on, keeps answers to the customer's instructions, tool results and knowledge articles, and the agent escalates when it lacks the information to answer. Coach audits the knowledge base for gaps and outdated articles and proposes the fixes, and knowledge sources such as Confluence, Guru, Google Docs and Notion connect directly.

Sourcelorikeetcx.airead 2026-11-02

Context carries within a conversation and across a voice call, and Fin passes full customer context to a human on escalation. No memory layer with a stated scope or lifetime is named, and the persistence Fin advertises is a learning flywheel that feeds conversations into model improvement, not a memory store that can be located or deleted for a given customer.

Sourcefin.ai/capabilitiesread 2026-09-05

Customer memory keeps context across every touchpoint, so a customer who starts in chat does not repeat themselves on a call, because the channels run on one engine. Dreamweaver goes further for each customer: it keeps that customer's goals, history and preferences, learns when they reply and how they like to talk, and acts on them over weeks and months. No scope, lifetime, retention window or per customer deletion control for that memory is described; the FAQ says a departing customer's data is deleted within 60 days of request.

Sourcelorikeetcx.airead 2026-11-02

Procedures carry deterministic controls and cross-team approvals, with multi-stage validation and approval workflows running before any action is taken, and Fin performs only actions the customer has explicitly configured and approved. Refund paths enforce eligibility windows, value thresholds and approval requirements, and Guidance bounds tone, scope and escalation.

Sourcefin.ai/trainread 2026-09-05

Guardrails run as checks separate from the agent itself, on both sides of the conversation. Agent guardrails review every reply before it is sent and message checks screen each incoming customer message, including for prompt injection, with up to ten of each per concierge; when one fires it can alert the team, make the agent rewrite its reply (two retries, then escalate) or escalate the ticket at once, and every trigger is logged on the ticket.

Underneath, customer isolation, server side identity validation, workflow scoped tool access and hard execution caps are enforced in code, and sensitive steps such as payments and identity checks run as code the agent cannot alter. People write the procedures the agent follows, handoffs arrive with the investigation done, Resolution Loop lets a person take over the same conversation thread, Coach makes changes only with approval, and Dreamweaver checks contact hours, limits and consent before every message and flags customers who need a person.

Sourcelorikeetcx.ai/product/trustread 2026-11-02

Intercom supports SSO with Okta, Azure AD and OneLogin, plus 2FA, SCIM provisioning and IP restrictions, and holds SOC 2, ISO 27001, ISO 42001, ISO 27701, AIUC-1 and HIPAA.

Sourcefin.airead 2026-09-05

Lorikeet is SOC 2 Type II audited and ISO 27001:2022 certified, with the reports on a public Vanta trust center at trust.lorikeetcx.ai, downloadable under NDA. It says it is HIPAA compliant and GDPR aligned, signs BAAs and DPAs, holds zero data retention agreements with all model vendors and does not train models on customer data; it does not hold PCI DSS, so card capture stays in the customer's own payment flow.

Controls include PII redaction configurable by category and applied automatically to restricted viewers, minimum necessary handling of protected health information, TLS 1.3 in transit and AES 256 at rest, no public internet access to production, row level tenant isolation and third party penetration testing. On the customer side, accounts are set up with user roles, role based access control limits which tools each team member can use through the MCP server, and helpdesk connections go through an admin approved authorization flow.

Sourcelorikeetcx.ai/product/trustread 2026-11-02

The Analyze stage includes Topics Explorer, CX Score quality scoring across every conversation, an Optimize Dashboard and configurable alerts, which measure what happened. No production trace showing why the agent acted as it did is described, and the view of Fin's reasoning sits inside Simulations, a test surface used before launch.

Sourcefin.ai/analyzeread 2026-09-05

Every customer interaction, model choice and action is tracked and explainable, not a black box. Every tool call, elevation request and reasoning step is logged with timestamps and can be replayed for a regulator examination, every guardrail event lands on the ticket and in guardrail metrics, and every configuration change made through the MCP server carries an audit trail.

Coach scores every AI handled conversation Good, Warning or Critical into a Ticket Quality Score, with human agent scoring on request, and answers analytics questions in plain language. Ticket history, outcomes and quality scores can be exported, and the FAQ lists audit trail exports, data access logs and compliance reports for regulatory reporting.

Sourcelorikeetcx.ai/product/trustread 2026-11-02

Customer data is hosted in the United States, the European Union or Australia, chosen to meet the customer's residency requirement, with redundancy across regions behind a 99.8% uptime SLA.

Sourcefin.airead 2026-09-05

Production runs on Google Cloud Platform in a private VPC with no public internet access and row level tenant isolation, encrypted with TLS 1.3 in transit and AES 256 at rest. Data storage residency is available in the US, Australia and the EU: the pricing page sets US residency on the Start and Scale plans, and Signature can choose a custom region. Lorikeet is delivered as SaaS; no self hosted or single tenant option is described.

Sourcelorikeetcx.ai/product/trustread 2026-11-02

Fin ships named role-specific agents that customers adopt as units, Fin for Service, Fin for Sales and Fin for Ecommerce, each with pre-trained skills, alongside packaged helpdesk editions for Salesforce, Freshdesk and HubSpot, a Simulation library and the AI Agent Blueprint.

Sourcefin.ai/capabilitiesread 2026-09-05

Each customer's configuration is written from its own standard operating procedures, by design, so time to value depends on the customer's procedure documentation; Coach builds the first concierge from a plain language description on a free trial.

There is no template library, prebuilt agent catalog or marketplace, and the specialist agents in the team of agents architecture work together internally rather than being assets a customer browses and adopts. Dreamweaver takes a goal such as retention or lifetime value with no SOPs, workflows or campaigns to write, which is a different route to the same end rather than a pack. Toolshed at lorikeet.tools is described as free CX tools and calculators.

Sourcelorikeetcx.airead 2026-11-02

Answers come from Fin's own proprietary model family, the Apex model built for customer support and Apex Flash for low latency voice, under the patented Fin AI Engine. No customer or admin model selection is described.

Sourcefin.ai/pricingread 2026-09-05

OpenAI, Anthropic and Baseten are among the core sub processors beside Google Cloud, so several model providers are in play, all under zero data retention agreements with no fine tuning on customer data, and Lorikeet routes between providers by task; on voice it switches models to balance latency and precision. The routing is Lorikeet's own and the model choice is tracked, not made by the customer, with no customer facing selection or bring your own model control.

Sourcelorikeetcx.ai/product/trustread 2026-11-02

The Fin API Platform exposes the Apex generative, Retrieval and Reranker models individually or as a complete RAG pipeline for building custom agents, and the Fin Agent API embeds Fin in a customer's own product without the Intercom messenger. API is a first-class deployment channel, and the versioned REST API comes with an OpenAPI specification, interactive reference and changelog. A help center note says Fin cannot serve conversations created through the API, but it predates the API Platform and covers only that one path.

Sourcefin.ai/api-platformread 2026-09-05

Outside callers can drive the platform three ways. The Lorikeet MCP server, released 11 March 2026 at a published endpoint (api.lorikeetcx.ai/v1/mcp), lets Claude, ChatGPT, Claude Code and Codex manage workflows, knowledge, tickets, tools, integrations, simulations, guardrails and analytics with read and write access, every change logged; its auth scheme and tool list sit in the developer documentation, which is behind a customer access code at docs.lorikeetcx.ai, so two of the four clauses for Full cannot be read.

A voice API lets a customer's phone system create a Lorikeet conversation with caller context before forwarding the call, and the SDK embeds Lorikeet in the customer's own product. Lorikeet also publishes public HTTP endpoints for outside AI assistants to ask its support agent questions, book demos and start a trial sign up. No SDK reference or API reference is readable without a customer login.

Sourcedocs.lorikeetcx.ai/guides/quickstartread 2026-11-02

Customers get a test suite covering Simulations, Regression Tests, Previews and Audience Testing, run against their own configured agent, with results readable before deployment.

Sourcefin.ai/pricingread 2026-09-05

Testing covers both sides of go live. Before launch, simulations generate scenarios and test customers from the customer's own historical tickets and synthetic cases, run them in batches against a draft concierge with tool responses mocked, and a pass rate threshold agreed before anything is published; adversarial scenarios such as prompt injection and false authority claims run the same way, suites rerun after every change with side by side batch diffs, and Coach can rerun failed scenarios and compare batches.

After launch, automated quality checks score every AI handled ticket Good, Warning or Critical into a Ticket Quality Score, human agent tickets can be scored once enabled, and Coach turns findings into workflow fixes that are tested in simulation and shipped with approval, with A/B experiments to compare approaches. A resolution the customer marks bad is not charged. Lorikeet also runs a public prompt injection challenge, Own Goal, against a live agent.

Sourcelorikeetcx.airead 2026-11-02

Every deployment surface is a messaging, voice or API channel, and actions on external systems run through helpdesk connectors and Data Connectors. No browser control, virtual desktop or operation of software without a programmatic interface is described.

Sourcefin.ai/pricingread 2026-09-05

Work happens programmatically inside the customer's systems through controlled integrations, APIs, webhooks and MCP servers, with tool calls such as retrieving recent transactions shown as named workflow steps. No browser control, hosted session or computer use is described, and Lorikeet's public HTTP endpoints let outside AI assistants call Lorikeet rather than operate any interface.

Sourcelorikeetcx.airead 2026-11-02