Chipp

Custom actions let an agent call any HTTP API mid conversation, with authentication, parameters, variables and a test panel, so a request such as adding a contact to a mailing list or opening a high priority ticket becomes a real API call. Built in integrations reach Gmail, Google Calendar, Drive and Sheets, HubSpot, Monday, Notion, Slack, GitHub, Dropbox, OneDrive and SharePoint, MCP pro actions add Asana, Fireflies, Otter and SeekOut, and database connections let an agent query a company's own data. Chipp sells one plan with no feature gates, so every customer gets the full action set.

Sourcechipp.ai/docs/custom-actions/overviewread 2026-10-18

Chipp Factory, which the documentation calls Alchemist, runs the full software delivery loop with its own agents: research, implementation, code review, documentation, deployment, verification, production monitoring and self repair, on every change from a plain language ticket.

Each stage hands its output to the next without a person finishing the work, and the monitoring loop investigates and repairs problems it can resolve safely in the deployed app. Chat agents built in the App Builder can be embedded into Factory projects, and agent behavior inside a conversation chains tools, knowledge and memory turn by turn.

Sourcechipp.ai/docsread 2026-10-18

Uploaded files, spreadsheets and URLs are extracted, split into chunks and embedded so the agent searches them by meaning rather than keywords, with RAG settings to tune retrieval. Enterprise knowledge bases scope shared collections to parts of an organization, Circles share files and agent memory across a group, and My Files keeps a document library shared across a user's assistants. Database connections let an agent query live business data alongside the indexed documents.

Sourcechipp.ai/docs/guides/how-knowledge-sources-workread 2026-10-18

Guardrails bound what agents do on their own: heartbeat outreach stays inside active hours, skips recently active users, stops after repeated empty checks, never repeats a message within a day and halts at daily or monthly budget caps. Access control sets who may chat, from open to an email allowlist, and custom actions are tested before use. Agents run their actions without a person approving each one. In Chipp Factory the code review stage is performed by its own agents, and no step asks a human to sign off before a change deploys.

Sourcechipp.ai/docs/guides/heartbeatread 2026-10-18

Organizations assign members to workspaces under four roles, with owners controlling billing and deletion, admins inviting members and managing workspaces, and members building inside assigned workspaces. An audit log on every plan records who did what and when for owners and admins, access control restricts who can use each app, and bring your own auth passes pre authenticated sessions from a customer's own login. HIPAA applies on every plan with HIPAA compliant models, backed by SOC 2 Type II and GDPR compliance, and independent penetration test findings are tracked publicly.

Sourcechipp.ai/docs/guides/team-managementread 2026-10-18

Every conversation is kept in chat history with transcripts available through the API, analytics show usage and engagement, a performance dashboard and programmatic usage stats track volume and cost, and message tagging and feedback mark answers for review. The immutable audit log records administrative actions with actor, time and resource, and session webhooks fire when an MCP integration loses its tools mid chat, so builders see both what agents said and what changed around them.

Sourcechipp.ai/docs/guides/audit-logsread 2026-10-18

User Memory extracts durable facts from conversations in the background and recalls them in later sessions: facts such as a user's company and role, preferences such as liking concise answers, standing instructions such as always replying in Spanish, and context such as an open evaluation. Each memory sits in a fact layer kept indefinitely or an episodic layer that decays over time, so the scope is per user and the lifetime is stated. Memory is on by default for new apps, and builders review and manage memories in the app or through the Builder API.

Sourcechipp.ai/docs/guides/user-memoryread 2026-10-18

Chipp is delivered as a hosted service, with white label and custom domains for how agents appear. There is no region choice, private cloud or self hosted option, and the pages for buyers do not name where data is hosted. Higher service tiers add a dedicated model capacity lane and bring your own model keys, which changes model routing but not where the platform runs.

Sourcechipp.ai/pricingread 2026-10-18

Templates are full agent blueprints with a system prompt, model, suggested messages and sometimes required integrations, and using one creates a working app in the workspace in minutes. Business Brain, Personal Assistant and Project Manager are the templates Chipp will configure on a customer's data for $2,500, and the template library also holds full stack app templates with live demos. Skills are smaller composable prompt pieces that add a capability to any agent.

Sourcechipp.ai/docs/guides/skills-and-templatesread 2026-10-18

Heartbeat lets an agent wake on its own and decide whether to message each user, within active hours, a cooldown after recent chats, a limit on consecutive empty checks, a 24 hour deduplication rule and daily or monthly budget caps, and it can deliver through chat or post JSON to a webhook. Users reach agents on websites and embeds, in Slack, Teams, WhatsApp and Discord, by email and text, and by phone, with outbound calling, and session lifecycle webhooks notify other systems when conversations start or integrations fail.

Sourcechipp.ai/docs/guides/heartbeatread 2026-10-18

Builders choose the model for each agent from OpenAI, Anthropic, Google and others, with guidance such as GPT 5.4, Claude Sonnet 4.6 or GPT 5.6 Terra for general work and faster or cheaper models for simple tasks, and HIPAA compliant models for regulated use. Organizations on the Build For You and Fractional CTO tiers can bring their own Google, OpenAI and Anthropic keys with no platform fee and run on a dedicated capacity lane apart from the shared pool.

Sourcechipp.ai/docs/guides/choosing-a-modelread 2026-10-18

A chat completions endpoint calls any Chipp app by its app ID with streaming, session continuation, session variables and MCP secrets, and further endpoints cover authentication, chat transcripts and organization export. The Builder API manages knowledge sources, memories, sessions, consumers, tags, analytics and feedback. The Chipp MCP server, added to Claude Code with one command and OAuth, lets Claude Code, Cursor or any MCP client list, read, create and manage Chipp apps.

Sourcechipp.ai/docs/api/referenceread 2026-10-18

Evals work as automated quality checks for an agent before it reaches customers: a builder defines test cases with the questions the agent must handle and the expected answers, runs them, and sees which responses match, such as catching an agent that quotes $99 a month when the real price is $29. Running the same set after a change to instructions, knowledge or model shows whether the agent got better or worse, and custom actions have their own test panel before they go live.

Sourcechipp.ai/docs/guides/understanding-evalsread 2026-10-18

Agents act through APIs, built in integrations, custom actions and MCP pro actions. The Desktop Agent, an Enterprise macOS menu bar app with screen context and a command bar, was retired in July 2026 and can no longer be installed. Nothing in the current product drives a browser or a desktop, so a site or application without an API is out of reach unless an MCP server the builder connects provides that reach. Chipp Factory verifies the apps it deploys as part of its delivery loop, but that checks Chipp's own builds and does not operate outside software on a customer's behalf.

Sourcechipp.ai/docs/guides/desktop-agentread 2026-10-18