MindPal

Composio connects an agent to more than 100 applications, among them Gmail, Slack, Google Calendar and Notion, through OAuth, and the builder chooses exactly which prebuilt actions each agent may take, so it can send an email, post a message or create a calendar event during a conversation. Any MCP server plugs in by URL to add its tools, and custom tools call any HTTP API with a chosen endpoint, method, headers and parameters, which covers pushing data to a CRM, updating a database or triggering another system.

Sourcedocs.mindpal.space/agent/composio.mdread 2026-10-18

Multi agent workflows chain agent nodes on a canvas with variables passing each output forward. Router and gate nodes branch on conditions, loop nodes repeat a step over a list, subflow nodes call another workflow, code nodes run custom logic, and an orchestrator worker node lets a lead agent split a task across worker agents and combine their results. Agents can also call sub agents mid conversation, and Mindie, a built in assistant, drafts workflows from a description.

Sourcedocs.mindpal.space/workflow/build/orchestrator-worker-node.mdread 2026-10-18

Knowledge sources give an agent the business's own material: websites plus uploaded Word, Excel, PDF, CSV, text and EPUB files of up to 200 MB each. Content is split into chunks, vectorized and stored, and the agent searches those chunks by semantic similarity whenever it needs information. The same sources can serve several agents, so a coach's course library or an agency's style guide is indexed once and reused. Plans include from 100 MB of knowledge storage on Free upward, and storage beyond the allowance costs $1 per GB a month.

Sourcedocs.mindpal.space/agent/knowledge-sources.mdread 2026-10-18

Supervised mode runs a workflow one step at a time and waits for a person to approve, review or edit each response before the next node starts. Human input nodes pause any workflow for a decision, extra context or feedback to the agents, and gate nodes hold the flow until a condition is met. Each Composio connection limits an agent to the actions the builder selected, and published agents can be rate limited per user.

Sourcedocs.mindpal.space/workflow/run/in-app.mdread 2026-10-18

Access splits into editor seats, for team members who build agents and workflows, and user seats, for clients who only use what is shared with them, with sharing controls on the Advanced plan and above. Data is encrypted in transit and at rest, uploaded content is never used to train models, staff access to customer data is restricted and logged, and enterprise customers can sign a data processing agreement. There is no single sign on, no finer grained roles and no named SOC 2 or ISO certification.

Sourcemindpal.space/data-securityread 2026-10-18

A workflow run shows each agent's output in real time, so a builder can see how every node contributed, and supervised mode stops at each step for inspection. Runs triggered by API return a run ID whose results can be fetched later or sent to a webhook, and the workflow preview traces the execution path when steps are skipped. Credit use per model is visible against the plan. There is no audit log of who ran or changed what, and no per run trace of tool calls.

Sourcedocs.mindpal.space/workflow/run/in-app.mdread 2026-10-18

Notes are text entries that people create and edit inside MindPal, and every agent with access reads them across conversations as a long term store of facts, preferences and business context. Audience Memory gives published chatbots and workflows a stable user ID so a returning visitor sees past conversations and resumes a saved run instead of starting over. The agent does not write its own memories from what it learns, and no retention period is set.

Sourcedocs.mindpal.space/agent/notes.mdread 2026-10-18

MindPal is a hosted service whose data lives in AWS S3 and MongoDB in US data centers only, encrypted in transit and at rest. There is no region choice, private deployment or self hosted edition, and no plan up to Ultra changes where data sits. Customers that need data kept in Europe or inside their own cloud have no option here beyond the data processing agreement offered to enterprise customers, which covers terms and not location. Published agents can appear on a customer's own website or custom domain, but they are embeds served from MindPal's US hosting.

Sourcemindpal.space/data-securityread 2026-10-18

A template library holds hundreds of ready made agents and multi agent workflows that a user copies into a workspace and adjusts, so a new builder starts from something that already runs. Mindie, the built in workflow assistant, drafts a new workflow from a plain description as another starting point, and published agents and workflows can be shared with clients through user seats or embedded on their own sites. Ultra adds agents and workflows built by MindPal for the customer, and MindPal Managed is a separate done for you service.

Sourcemindpal.spaceread 2026-10-18

Schedule triggers start a workflow at recurring times, such as weekly content planning or monthly reports, and the public API trigger starts one from any outside system with a POST request, which is how Zapier, Make and custom apps hand events to MindPal. Workflows also run from a public form, inside the app or in the background, and agents answer through chatbots embedded on websites and on platforms such as Kajabi, Circle and GoHighLevel.

Sourcedocs.mindpal.space/workflow/trigger/schedule.mdread 2026-10-18

Each agent runs on a model the builder picks from OpenAI's GPT 5, GPT 5 Mini and o series, Anthropic's Claude Opus, Sonnet and Haiku, Google's Gemini 3.0 Pro and 2.5 models, DeepSeek V3 and R1, xAI's Grok and Llama and Kimi on Groq. Models cost different credit amounts per run, from half a credit for DeepSeek to one credit for Gemini 2.5 Flash and three for GPT 5 Mini, so builders can put cheap models on simple steps and premium ones on hard reasoning.

Sourcedocs.mindpal.space/agent/language-model.mdread 2026-10-18

A public REST endpoint starts a workflow with custom input data and returns a run ID, and a results endpoint or a webhook node at the end of the workflow delivers the output, which lets other software call MindPal workflows. API access comes with the Advanced and Ultra plans. Custom session context and user IDs pass data into published agents. The API covers running workflows only, while creating or managing agents, knowledge and workflows happens in the app, and there is no SDK or MCP server for MindPal itself.

Sourcedocs.mindpal.space/workflow/trigger/api.mdread 2026-10-18

The evaluator optimizer node has one agent produce output and an evaluator agent review it against set criteria and send feedback, repeating until the result satisfies them, which raises quality inside a single run. Supervised mode and debugging guides help a builder test steps by hand. There is no saved set of test inputs, expected outputs or scores for comparing a changed agent or workflow against the previous version.

Sourcedocs.mindpal.space/workflow/build/evaluator-optimizer-node.mdread 2026-10-18

Agents act through custom HTTP API tools, Composio actions and MCP servers. None of the tools drives a browser or desktop, and websites enter MindPal as knowledge sources that are crawled and indexed ahead of time, then searched while the agent works. A workflow that needs data from a site with no API can only reach it through an MCP server that supplies browsing, attached by the builder. Embedded chatbots run inside a visitor's browser on the customer's site, but that is where people talk to the agent, and the agent never clicks or types in a page itself.

Sourcedocs.mindpal.space/agent/tools.mdread 2026-10-18