Ideavo

Predefined MCP servers switch on with a toggle, among them Neon, Supabase, Context7, Kaggle, Firecrawl, Playwright and Rube for third party apps through Composio, and any other MCP server can be added by URL or configuration at project or global scope. One click integrations provision a Neon PostgreSQL database with Drizzle and Better Auth sign in for Next.js projects, the agent wires in Auth0, Clerk or other providers for other stacks, Razorpay adds checkout and subscriptions, and GitHub, Vercel, Railway and E2B connect from the project bar, with keys held in a Secrets tab.

Sourcedocs.ideavo.ai/docs/features/mcpread 2026-10-18

Build Mode works autonomously by default, exploring the codebase, applying coordinated changes across frontend, backend and configuration, and iterating until the result works. Tasks split a project into parallel workstreams, each in its own isolated sandbox with its own conversation, tracked on a Kanban board from Draft to Merged, and a merge flow rebases on the default branch and hands any conflicts to the agent. Queued prompts keep running one after another even after the builder leaves the project.

Sourcedocs.ideavo.ai/docs/features/tasksread 2026-10-18

Up to five files of 20MB each, including screenshots, documents and PDFs where the chosen model reads them, can be attached to a prompt, and @ mentions point the agent at specific files in the project. Existing GitHub repositories import with no manual setup, Build Mode runs real time research on the web, and the Context7 MCP adds library documentation. There is no built in knowledge base for a team's own documents beyond attachments and whatever MCP servers are connected.

Sourcedocs.ideavo.ai/docs/features/attachmentsread 2026-10-18

Plan Mode is read only, so the agent can investigate and propose a plan before anyone switches to Build Mode, and when it needs a permission or an answer it pauses and asks. Every assistant message carries a version that can be restored or forked, task changes reach the main version only through a person's merge, and roles keep Viewers out of the chat, IDE and secrets. Build Mode itself has no approval setting, so changes inside a run apply without a confirmation step.

Sourcedocs.ideavo.ai/docs/features/restoreread 2026-10-18

Ideavo holds SOC 2 certification and encrypts data at rest, with audit trails and SSO for teams, and enterprises get SSO, SCIM and pooled usage across accounts. Projects have Owner, Admin, Editor and Viewer roles with access requests defaulting to the lowest level, API keys and OAuth tokens are encrypted and cannot be viewed after saving, and secrets are masked, never committed and cannot be attached to prompts. No SOC 2 report, trust portal or SSO setup guide is offered publicly.

Sourceideavo.airead 2026-10-18

A Details view follows each run step by step, showing the current step, the files being read or changed and the tools in use, such as web search, log inspection and verification. Every assistant message is a numbered version, GitHub keeps the commit history, and running the sandbox on the builder's own E2B account shows compute use and billing there. Teams get audit trails, but no export of agent activity or admin usage dashboard is offered.

Sourcedocs.ideavo.ai/docs/features/build-moderead 2026-10-18

An AGENTS.md file in the project holds coding style, preferred tools, project structure and communication preferences, and the agent loads it in current and future sessions. Installed skills apply across projects, each task keeps its own conversation history, and long conversations compact once a model's context limit is known. Restoring a version also restores the rules and the agent's context, but no memory of preferences learned from past chats exists outside those files.

Sourcedocs.ideavo.ai/docs/features/agent-rulesread 2026-10-18

Apps deploy to the builder's own Vercel or Railway account with a live production URL and custom domain, mobile builds follow Expo's route to the app stores, and production runs with its own secrets and database separate from the sandbox. Code exports as a ZIP or syncs to GitHub, and the sandbox can run on the builder's own E2B account. Ideavo itself is a hosted service with no self hosted edition or region choice.

Sourcedocs.ideavo.ai/docs/features/publishread 2026-10-18

Community skills from skills.sh install into the agent with a single command in the IDE terminal, giving it procedural knowledge for areas such as frontend design, testing or data pipelines across every project. A curated set of predefined MCP servers is ready to switch on. There are no starter templates or packaged agents for particular kinds of app; projects begin from a prompt or an imported repository.

Sourcedocs.ideavo.ai/docs/features/skillsread 2026-10-18

Work starts from a prompt in the project chat on the web or on a phone, where the Kanban board is also available, and follow up prompts can be queued, paused or cleared while the agent is busy. Draft tasks can be staged with a title and prompt for a teammate to start later. Nothing starts the agent on a schedule, from a chat channel or from an outside event, so each run begins with a person's prompt.

Sourcedocs.ideavo.ai/docs/features/prompt-queueread 2026-10-18

Builders bring their own API keys or subscriptions from Anthropic, OpenAI, Google, OpenRouter, OpenCode, Kimi and other providers, including ChatGPT Plus or Pro through OAuth for subscription models such as Codex, and pick the model per request with no markup. Custom providers add local models through Ollama, self hosted endpoints or any OpenAI compatible base URL with their own model lists and limits, and an effort setting controls how much reasoning a model uses.

Sourcedocs.ideavo.ai/docs/features/byokread 2026-10-18

No API, SDK, command line tool or MCP server lets an outside tool create projects, prompt the agent or deploy an app, so work goes through the Ideavo interface. Code leaves through GitHub sync and ZIP export as a standard repository that other tools can work on, and the agent itself is extended from inside through AGENTS.md rules, skills, custom MCP servers and custom model providers.

Sourcedocs.ideavo.ai/llms.txtread 2026-10-18

Build Mode checks its own work: it watches build errors, inspects console output and runtime logs, reproduces reported bugs to confirm the cause before fixing, and iterates until behavior matches the request. Plan Mode handles deep debugging and security or performance reviews without touching code, the Playwright MCP adds visual and end to end tests of the running app, and the preview emulates desktop and popular phone sizes.

Sourcedocs.ideavo.ai/docs/features/build-moderead 2026-10-18

With the predefined Playwright MCP switched on, the agent drives a real browser in the sandbox to click through, submit forms and navigate the app it is building, and to fetch pages that need JavaScript to render, while Firecrawl handles scraping and extraction from websites. That browser control is aimed at testing the project and reading pages. The agent does not sign in to outside sites or carry out tasks there for the builder, and there is no desktop control.

Sourcedocs.ideavo.ai/docs/features/mcpread 2026-10-18