Flozic

Agents read and write across more than 1,000 app integrations, among them HubSpot, Salesforce, Gmail, Slack, Notion, Stripe, Shopify, Google Sheets, QuickBooks, NetSuite and GoHighLevel, each connected with a one click OAuth sign in instead of pasted API keys. When no connector covers a system, an agent calls any webhook, REST API or custom HTTP endpoint directly. Typical actions are creating and updating CRM records and deal stages, sending email, posting to Slack, reconciling payments in Sheets and filing tickets in Zendesk, Freshdesk or Jira.

Sourceflozic.airead 2026-10-18

A plain English goal becomes an agent with its trigger, LLM prompt and action nodes, which a team then refines on a visual canvas where every step can be seen, tested and edited. Agents plan their steps from context instead of one fixed path and work through conditional branches, loops, filters, formatters and delays, so a lead that falls outside the usual route still gets handled. Retry policies can be set per step, Professional and above replay tasks that failed on temporary errors, and one flow can pass screen work to a computer use agent and the rest to API connected agents.

Sourceflozic.airead 2026-10-18

Agents are trained on a team's own documents, websites and help center, along with connected knowledge bases such as Notion, and answer through retrieval augmented generation that cites its source, for example replying in a Slack thread with the answer and a citation.

Conversational agents resolve tier one support questions from the knowledge base and escalate when confidence is low, and autonomous agents draw on the same knowledge to draft support replies. Supported file types, refresh of connected sources, size limits and retrieval settings go unstated, so grounding depth past cited answers is unclear.

Sourceflozic.ai/agents/conversationalread 2026-10-18

Approval steps can sit anywhere on the canvas so nothing sensitive runs without a person's tap, and each agent's access can be scoped.

Computer use agents show their first run live for approval and gate sensitive actions behind a human, conversational agents hand off to a person with the full transcript when confidence is low, and agents flag edge cases instead of guessing.

Voice agents call only allowed numbers under a daily call limit, skip do not call entries and keep recording off unless switched on, with a consent message. How approvals reach a reviewer, and what happens when one goes unanswered, is not set out.

Sourceflozic.ai/agents/autonomousread 2026-10-18

Enterprise security covers SOC 2 Type II and GDPR compliance, role based access control, single sign on through SAML or OIDC, and audit trails of every agent decision, and parent Appy Pie holds a SOC 2 attestation along with ISO 27001:2013 and ISO 22301:2019 certifications. Data at rest is encrypted with keys managed in AWS Key Management Service.

App connections use scoped OAuth access with each tool's policy enforced centrally, computer use logins stay in an encrypted vault used only inside a sandbox, and customer data is not used to train models. Enterprise adds PCI DSS compliance with HIPAA and HITECH eligibility.

Sourceflozic.airead 2026-10-18

Each run lands in a live run history with step by step inputs and outputs, the decisions taken, error traces and timing, and failures can raise alerts in Slack or by email. Conversational agents log every conversation with its inputs, outputs and latency for replay and tuning, and voice agents enter each call and its outcome in a Call Ledger. Computer use agents narrate their plan, show each on screen action as it happens and keep a full recording of every run to replay, debug and audit. Audit trails of agent decisions are part of the enterprise security set.

Sourceflozic.airead 2026-10-18

Conversational agents keep multi turn memory over a conversation's full history, and a handoff to a person carries the full transcript along with the CRM record. Autonomous agents keep a log of each run's inputs, decisions and outputs, which people use for review and tuning and which later runs do not read back. There is no separate memory store that an agent writes facts to and edits over time, so durable state about a customer or a case lives in the CRM records, sheets and Notion pages the agent updates as it works.

Sourceflozic.ai/agents/conversationalread 2026-10-18

Flozic is a hosted service on Appy Pie's AWS infrastructure, with no self hosted, private cloud or customer tenant edition. Enterprise customers get VPC peering and on premise connectors, which let agents reach databases and apps inside a private network, but the platform, its run logs and its agent data stay in the vendor cloud.

The Enterprise plan adds a choice of data retention options, which governs how long data is kept and not where. Customers cannot pick a data region, so teams with residency rules have no way to pin where agent data is stored, and voice agents run over the customer's own Twilio account.

Sourceflozic.airead 2026-10-18

The agent library holds 29 ready made autonomous agents, 28 conversational agents and 26 voice agents, across CRM and sales, marketing reporting, finance and invoicing, support triage, HR onboarding, ecommerce, content and social, plus receptionist, outbound sales, booking and collections agents on the phone. Each agent type's prompt box offers starter prompts, such as HubSpot lead enrichment posted to Slack, Stripe payment reconciliation in Sheets or a daily billing portal download, that fill the builder for editing, and conversational templates cover support, sales, FAQ and lead generation.

Sourceflozic.ai/agentsread 2026-10-18

Autonomous agents start from instant push triggers on app events such as a new Gmail message, a new Airtable row, a Stripe payment or a Slack message, and also from polling, form submissions, custom webhooks and cron schedules, with run intervals of 5 minutes on Standard, 2 on Professional and 1 on Business.

Conversational agents take messages from a web widget, WhatsApp, Slack, Microsoft Teams, Telegram, email and SMS under one configuration, and voice agents answer inbound calls and place outbound ones on the customer's own Twilio number. Computer use agents run on a schedule, a webhook or an app event.

Sourceflozic.ai/agents/autonomousread 2026-10-18

Each agent runs on the model its builder picks, from GPT, Claude, Gemini, Llama and Mistral, and bring your own key lets a team put its own provider account behind an agent. Computer use agents need models that can see, so they run on the vision capable GPT, Claude and Gemini models, chosen per task. Inside a flow, action steps can hand a classification, summary or decision to GPT, Claude, Gemini or another LLM, so one agent can mix models across steps. No single lab's model is required, and the choice is made agent by agent.

Sourceflozic.ai/agents/autonomousread 2026-10-18

Outside systems start Flozic agents through custom webhooks, and that is the one way in from code. A universal MCP server at one endpoint, mcp.flozic.ai, lets Claude, ChatGPT, Cursor, Windsurf or any MCP client reach more than 2,000 MCP servers and Flozic's 1,000+ apps as tools, with scoped OAuth access per app under centrally enforced policy.

That endpoint hands an outside assistant the apps Flozic connects to; it does not create, edit or run Flozic agents. There is no SDK or public API reference for building or managing agents in code, so a developer works through webhooks, the MCP bridge and HTTP calls from inside flows.

Sourceflozic.ai/mcpread 2026-10-18

Triggers can be tested with live sample data before an agent goes live, every step on the canvas can be tested and edited, and a computer use agent's first run plays live, step by step, for approval before it goes on a schedule. Agents can be paused, versioned, forked or rolled back in one click, and run history with inputs, outputs and error traces supports replay and tuning after the fact. There is no test suite, dataset evaluation or scoring of agent outputs before a change ships, so quality checks rest on a person watching runs and reading the logs.

Sourceflozic.airead 2026-10-18

Computer use agents read the screen with vision models, then click, type and navigate websites, internal portals and legacy desktop apps that have no API, planning the on screen steps from a plain English description and adapting when the interface changes.

They work inside an isolated sandbox with logins drawn from an encrypted vault, can be watched live, and record every action for replay. Typical jobs include bulk record entry into a browser CRM, researching company websites into a sheet and downloading invoices from a billing portal, each set to run on a schedule, a webhook or an app event.

Sourceflozic.ai/agents/cuaread 2026-10-18