Crisp
Also known as: crisp.chat, Crisp Hugo, Hugo by Crisp, hugo.ai
Crisp combines Hugo, an AI customer support agent, with a shared inbox, knowledge base and workflow builder. Hugo handles incoming conversations, retrieves approved content and calls connected tools for account or order tasks. Teams can choose among several model providers, set escalation rules and compare agent variants before changing production. Plus adds detailed execution traces and reviewed knowledge suggestions. Paid plans include an AI usage allowance, with optional additional consumption under a spending limit.
Crisp IM SAS operates a customer messaging and support platform with Hugo as its embedded AI agent. Hugo works across website chat, email and connected messaging channels, using business knowledge and tools to resolve requests or pass them to employees. Teams choose the channels and office hours where the agent runs, decide how reopened conversations are handled, and direct particular topics into a workflow or team inbox. The support agent sits alongside shared customer records, human conversation handling and a knowledge base.
Hugo can call external MCP servers with bearer token or basic authentication. Those tools can update a record, create a ticket or process an order, while native Shopify, PrestaShop, WooCommerce and Stripe integrations retrieve account, order or invoice information. Sensitive tools can require customer verification by an email or SMS code. Workflows add branching, field collection, customer data updates, webhooks and transitions back to Hugo. Customers can begin with a template or import a JSON workflow.
Widget Tools extend actions into the visitor's current web session through JavaScript handlers supplied by the customer's developers, for example opening an upgrade flow or applying a coupon. They require the live chatbox and do not run in simulations. Knowledge comes from crawled sites, uploaded files, private question and answer snippets, and Crisp help articles. Help article changes synchronize automatically; external websites require a manual refresh.
Mini costs $45 per workspace per month and includes four seats and $5 in monthly Hugo credits. Essentials costs $95 with ten seats and $25 in credits; Plus costs $295 with at least twenty seats and $75 in credits. A 14 day trial exposes the feature set. AI consumption depends on tokens, context and model choice, with optional additional usage and a spending limit. Without that option, exhausted credits cause conversations to escalate to people.
Workspace access separates Owners and Members and supports enforced two factor authentication. Crisp has not undergone a SOC 2 audit. Messaging and plugin data reside in the Netherlands and Germany. Supported models offer an EU processing setting, and Crisp's private inference option runs on its own infrastructure in Paris rather than in the customer's environment.
The platform fits support teams that want automation and human work in one inbox. Customers select OpenAI, Anthropic, Google, Mistral or a Crisp hosted model, with a hosted fallback for provider failures. Playground comparisons run agent variants against the same question. Plus exposes chronological tool traces and AI Suggestions that propose knowledge changes for staff to approve. Hugo uses the active conversation context, while workflow variables last through a run. Its browser actions depend on registered application functions, without general screen navigation or desktop control.
Vendor details
Canonical URL
https://crisp.chat
Category
Customer support agent
Subcategory
Hugo support agent with MCP actions, visual workflows and shared inbox
Funding status
Crisp IM SAS, registered at 2 boulevard de Launay in Nantes, France. Created in 2015 and self funded with no venture investors, it reports more than 600,000 users.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Native MCP integrations retrieve Shopify, PrestaShop, WooCommerce and Stripe records. Custom MCP servers support authenticated actions in customer systems. Workflows call webhooks, update customer data and hand conversations to Hugo. Widget Tools invoke customer supplied frontend handlers. REST endpoints and the Chatbox SDK can start agents and workflows.
In practice
An online store connects an order lookup and a custom cancellation tool; Hugo verifies the customer before using the sensitive action.
A software company routes an upgrade request into a workflow that collects the account details and opens the upgrade interface through a registered Widget Tool.
A support manager reviews a proposed knowledge correction against the conversations behind it, accepts the relevant edits and saves the updated article.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
11.5 / 14 capabilities · 82%
| Integrations & Tool Calling | Full |
|---|---|
|
Authenticated MCP connections let Hugo call tools that create tickets, update records and process orders in customer systems. Connections support bearer tokens or basic authentication, with optional signed requests. Sensitive tools can require an email or SMS verification code before execution. Native Shopify, PrestaShop, WooCommerce and Stripe connections supply account, order and invoice lookups; custom servers define additional actions. Sourcedocs.crisp.chat/guides/hugo/mcp-api/quickstartread 2026-10-20 |
|
| Workflow Orchestration | Full |
|
The workflow builder connects events, conditions, actions and agent blocks into branched support processes. A workflow can collect information, update customer data, call a webhook and hand the conversation to Hugo. Routing rules let Hugo enter the appropriate workflow when it detects a situation. Exit blocks can start another workflow, while proxy blocks create loops that return to an earlier part of the process. Sourcehelp.crisp.chat/en/article/getting-started-with-workflows-hrmysbread 2026-10-20 |
|
| Knowledge Grounding & RAG | Full |
|
Stored knowledge combines website pages, uploaded TXT, PDF or CSV files, and private question and answer snippets. Crisp help articles synchronize with Hugo when created or changed, and disappear from its resources when deleted or unpublished. External website content needs a manual refresh, and the crawler reads static content without executing JavaScript. Builders can filter page paths and select which resource types the agent uses. Sourcehelp.crisp.chat/en/article/how-to-train-hugo-ai-agent-on-your-data-ijkdrlread 2026-10-20 |
|
| Human Oversight & Guardrails | Partial |
|
AI Suggestions, on Plus, proposes additions, edits and removals to the knowledge Hugo uses, links each recommendation to the customer conversations behind it, and saves nothing until staff accept or reject each change. At runtime, routing and escalation settings send sensitive topics, frustrated users or unanswered questions to a human inbox, and sensitive tools can require the customer to confirm with an email or SMS code. Hugo's replies and tool calls themselves go out without a staff approval step. Sourcehelp.crisp.chat/en/article/how-to-use-hugo-ai-suggestions-to-improve-customer-support-knowledge-ho0flfread 2026-10-20 |
|
| Security, Identity & Governance | Partial |
|
Workspace roles split Owners, who administer sensitive settings, from Members, who work in the inbox without critical configuration access, and only an Owner can change a role. Owners can stop other operators from deleting conversations or contacts, two factor authentication can be enforced for the whole workspace, and private inbox membership limits who sees selected conversations. On compliance, Crisp states it has not been through a SOC 2 audit and describes itself as committed to aligning with ISO 27001, without a certificate, alongside encrypted transport and restricted staff access. Sourcehelp.crisp.chat/en/article/how-to-be-legally-compliant-to-your-customers-with-crisp-1ih9v85read 2026-10-20 |
|
| Observability & Auditability | Full |
|
Hugo Observability on Plus shows a chronological trace of the actions between each customer message and answer. Expandable steps expose tool names and providers, input parameters, returned values, request and response logs, execution time and status. Traces also include searches, authentication, guardrails and workflow steps. Operators can filter failed or escalated conversations, open the original inbox thread and export one or all available traces. Sourcehelp.crisp.chat/en/article/how-to-monitor-and-debug-hugo-ai-agent-conversations-with-observability-1q79tp1read 2026-10-20 |
|
| Memory & State Persistence | Partial |
|
Workflow Memory stores submitted field values under named keys for use in later messages, conditions or profile updates. API injected variables belong to the workflow run, persist across its steps and can be overwritten by memorization blocks. Hugo itself answers from the ongoing conversation and its configured knowledge resources. Details that should outlast a conversation, such as an email or a plan, are written to the contact's profile fields, which operators and workflows read later; Hugo keeps no separate store of what it learned about a customer. Sourcehelp.crisp.chat/en/article/how-to-ask-users-for-information-and-memorize-them-with-workflows-1oaygcread 2026-10-20 |
|
| Deployment & Data Residency | Full |
|
A Region setting next to the model selector lets a workspace process Hugo's requests globally by default or in the European Union, where the chosen model supports it, and the model comparison table lists the hosting regions each model offers. Platform messaging data is stored in the Netherlands and plugin data in Germany. For teams that want no external model API at all, Crisp runs private model inference on its own infrastructure in Paris, managing capacity and model updates itself; nothing installs in the customer's own cloud. Sourcehelp.crisp.chat/en/article/how-to-choose-and-configure-hugos-ai-model-1m4mm57read 2026-10-20 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
A library of chatbot and AI agent workflow templates covers third party data push, HTTP calls, conditional logic, website and product behavior, custom answers and automated routing, and the builder's New Workflow menu offers Template workflow beside a blank canvas and JSON import. Customers adopt a template, then add or change its event, action and condition blocks before deploying it for Hugo routing or automatic message triggers. Help center guides add worked examples such as an out of office or weekend responder that points visitors to self service resources until the team is back. Sourcecrisp.chat/en/chatbotread 2026-10-20 |
|
| Triggers & Channel Coverage | Full |
|
Hugo automatically handles incoming conversations on the channels enabled for the workspace, including chat, email, Instagram and WhatsApp. Office hour controls restrict when the agent runs, and a separate setting governs replies after a resolved conversation reopens. Workflows can start on new messages or through server API and Chatbox SDK calls, allowing an application event to launch the process before the visitor sends a message. Sourcehelp.crisp.chat/en/article/how-to-navigate-the-hugo-interface-and-features-h1eatvread 2026-10-20 |
|
| Model Flexibility & Routing | Full |
|
Automatic selection picks the model for the workspace, currently OpenAI's GPT 5.6 Luna, and manual selection lets a team choose among OpenAI GPT, Anthropic Claude, Google Gemini and Mistral families or a model hosted by Crisp. A comparison table rates each option on intelligence, relative credit usage and available hosting regions, and a built in fallback switches to Crisp's hosted model when the chosen provider fails. Hugo does not accept a customer's own model API key or an arbitrary model endpoint. Sourcehelp.crisp.chat/en/article/how-to-choose-and-configure-hugos-ai-model-1m4mm57read 2026-10-20 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
The REST API v1 reference covers the platform's own objects, including websites, conversations, people profiles, operators, verify settings and audit log entries, alongside an RTM API for real time events and a web hooks reference. For the agent specifically, the Workflow API lets an authenticated backend start Hugo or a named workflow for a Crisp session; an agent trigger can carry a custom instruction and a workflow trigger accepts variables for later steps. The Chatbox SDK starts a workflow from the customer's website, and the builder generates request snippets in Node, Python, PHP and Ruby with the workflow identifier filled in. Sourcedocs.crisp.chat/references/rest-api/v1read 2026-10-20 |
|
| Testing, Debugging & Optimization | Full |
|
The Playground under AI Agent, Evaluate tests Hugo with the current settings before a change goes live. Its Compare model options view runs several agent variants side by side on the same customer question, each on a different model, so a team can judge answer quality, tone, speed, credit usage and source grounding before choosing the configuration for production. The help center recommends drawing test questions from real conversations, including follow ups, complex cases and unclear requests. Comparisons are read by a person; the Playground keeps no saved test set and produces no automatic score. Sourcehelp.crisp.chat/en/article/how-to-choose-and-configure-hugos-ai-model-1m4mm57read 2026-10-20 |
|
| Browser / Computer-use | Not documented |
|
Widget Tools let Hugo ask the visitor's chatbox to execute a registered JavaScript function, such as opening an upgrade flow or applying a coupon. The handler runs in the customer's existing web session and returns a result to the agent. These actions depend on a developer supplied programmatic interface; Hugo has no general browser or desktop controller for navigating screens, finding buttons or operating applications without such handlers. Sourcedocs.crisp.chat/guides/hugo/widget-toolsread 2026-10-20 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
$45 per workspace per month
Workspace subscription plus token based Hugo consumption
Included quota
Mini: four seats and $5 monthly AI credits, around 90 automated conversations. Essentials: ten seats and $25, around 450. Plus: twenty or more seats and $75, around 1,350. The free plan has two seats and no AI credits.
What is public
Workspace prices, seats, monthly AI credits and the trial.
Billing mechanics
Subscription plus AI token consumption; monthly unused credits expire.
Cost watchouts
Paid integrations, additional contacts, seats beyond included limits and AI usage can add charges.
Variable cost rationale
AI use varies with model and conversation context, but additional consumption is optional and budget limited.
Additional watchouts
Workflows and advanced AI operations require higher plans; trial testing can consume credits.
Overage / add-ons
Optional additional usage has a customer spending limit. Without it, exhausted credits trigger human escalation.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free inbox with no AI credits; 14 day trial of the paid feature set
Lowest paid plan
Mini
Commercial notes
Annual payment is available. Prices exclude applicable VAT.
Key ambiguities
The actual AI cost per conversation depends on the selected model and workload.
Cancellation / refund
Monthly plans have no commitment period; termination removes plan access and the current month is nonrefundable. Annual terms govern the renewal period.
Support SLA / resale
Enterprise includes custom onboarding and a personalized SLA.
Missing data
An exact cost for a particular conversation workload.
Related vendors
- Decagon — AI agent platform purpose-built for customer support, with deep…
- 5.Y — 5
- Ada — Agentic customer experience platform whose AI agents autonomously…
- Aisera — Enterprise agentic AI platform that orchestrates specialized agents…
- ASAPP — Generative AI for enterprise contact centers that plugs into…
- Assembled — AI customer support orchestration platform that unifies autonomous…