Waniwani

Documented custom tool support that acts in outside systems. The SDK compiles the customer's tools, flows and widgets into an MCP server; flow nodes and api/ endpoints call the customer's own services (quote engines, booking calendars, lead handoff to a CRM or webhook), and the Email module sends email from a tool or flow node with one call, logged per conversation.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Tools, Endpoints, Lead generation guide, Email module)read 2026-09-25

A documented workflow model. createFlow builds a typed state graph (Zod typed state, named nodes, direct and conditional edges, interrupts, widget signals) that compiles to one MCP tool, with deterministic step order, validation and re-ask on failure, auto skip of filled fields, and state that resumes across tool calls. The docs place it against LangGraph as a funnel shaped graph.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Flows, Waniwani vs LangGraph) and llms.txtread 2026-09-25

A maintained retrieval structure over the customer's own content. The Knowledge Base ingests the customer's markdown (with metadata), stores it in Waniwani's hosted KB service, and searches it with embeddings; the distribution template registers a search tool on top of it, with configurable not found and refusal text for regulated corpora. The homepage describes natural language answers sourced from the customer's documentation.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Knowledge Base overview, App config search options) and the homepageread 2026-09-25

Guardrails and gates, with no person approving an agent action. Compliance Monitoring runs real time checks on mandatory disclosures, legal text and advice boundaries, with per jurisdiction rules and gate enforcement for multi step quoting and escalation workflows; the knowledge base search lets a regulated app name the human to go to when nothing matches. A flow interrupt asks the buyer a question, not a reviewer.

SourceWaniwani, the homepage (Compliance Monitoring) and docs.waniwani.ai llms-full.txt (Flows, App config); waniwani.airead 2026-09-25

Access controls and a compliance posture are both published. Access is governed by organization roles and membership, OAuth scopes checked server side on every API call, and admin issued service account keys with chosen read and write scopes per area and a separate toggle for destructive operations, revocable at once. The site's llms.txt states SOC 2 Type II Certified, alongside GDPR compliance, and no first party statement contradicts it.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Service accounts, MCP server How it works) and www.waniwani.ai/llms.txt (Compliance and Security)read 2026-09-25

Run level visibility. AI App Analytics lets a team drill into individual sessions with full tool call logs, alongside funnel volume, conversion, lead capture and drop off per platform; the SDK's tracking sends typed events tied to a session id derived from MCP metadata, with external user ids attachable, and the Waniwani MCP server returns session breakdowns and analytics digests.

SourceWaniwani, the homepage (AI App Analytics) and docs.waniwani.ai llms-full.txt (Tracking, Sessions, Identify users, MCP server); waniwani.airead 2026-09-25

State inside one conversation's flow. The flow engine keeps state in a KV store keyed by the MCP session, hosted by Waniwani or self hosted through adapters, so a funnel resumes across tool calls without passing state through the model; the chat widget can keep thread history on the user's own device. That is state inside a single session; no memory the agent carries across sessions with a stated scope and lifetime is documented.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Flows, KV store, Chat widget)read 2026-09-25

A named region list with a selection surface. Waniwani runs two independent stacks, EU and US, each with its own database, dashboard and MCP server; the customer picks the region at organization creation and the organization's data never crosses to the other. The flow state store can also be self hosted through KV adapters, and the SDK is MIT licensed code the customer runs.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Regions and data residency, KV store)read 2026-09-25

Assets the customer assembles. The managed project template ships a ready made knowledge base search tool, and the docs give funnel recipes (sales, lead generation, booking, insurance and pricing quotes) and a Kit with widgets and flows to build from; none is a complete agent a buyer selects and runs as shipped.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Kit, Build a funnel guides, App config)read 2026-09-25

Work arrives through several inbound channels, and nothing documented starts the app on its own. An app built on Waniwani is reached wherever a buyer asks: ChatGPT, Claude, Gemini, Perplexity and the customer's own website through the embeddable chat widget. Each is an assistant or a person invoking the app, which is channel breadth rather than an event, schedule or webhook starting work.

SourceWaniwani, www.waniwani.ai/llms.txt (AI Platforms Supported), docs.waniwani.ai llms-full.txt (Chat widget) and the homepageread 2026-09-25

No model choice. An app built on Waniwani runs inside the buyer's assistant, so the host's model does the reasoning; being reachable from ChatGPT, Claude or Gemini is not a choice of model. The chat widget talks to Waniwani's hosted chat backend with no model named or selectable.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Chat widget, What is Waniwani?)read 2026-09-25

Outside callers drive the platform. Waniwani's own hosted MCP server publishes endpoints per region (mcp.waniwani.ai/mcp, eu.mcp.waniwani.ai/mcp), signs callers in with OAuth 2.1 and scopes, and offers four tools (whoami, set_active_org, search, execute), where execute runs typed platform operations that create environments and API keys. Alongside it: the MIT licensed SDK (the WaniWani-AI/kit repository is MIT), the platform CLI, API keys and scoped service account keys.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Waniwani MCP server, How it works, Service accounts, CLI) and GitHub WaniWani-AI/kitread 2026-09-25

An evaluation loop on the customer's deployed app. Top of Funnel Monitoring sends synthetic buyers through every major AI platform the way real customers would, verifies disclosure accuracy, pricing and conversation quality per platform, runs regression tests on every model update and alerts on changes; Compliance Monitoring adds cross model consistency scoring; persona A/B testing is listed. The platform's scopes include creating, editing and running evaluations and reading evaluators and their results.

SourceWaniwani, the homepage (Top of Funnel Monitoring, Compliance Monitoring), www.waniwani.ai/llms.txt (Synthetic Buyer Monitoring) and docs.waniwani.ai llms-full.txt (Service accounts scopes); waniwani.airead 2026-09-25

No agent operates a browser or desktop. Widgets render inside the host assistant and the chat widget embeds in the customer's own site; that is the product appearing in an interface, not an agent navigating one.

SourceWaniwani, docs.waniwani.ai llms-full.txt (Widgets, Chat widget)read 2026-09-25