Mindra

Apps connect by OAuth from a connectors catalog the vendor puts at 3,000+, custom MCP connectors take an HTTPS URL with optional OAuth, and REST or OpenAPI tools are registered with authentication and operations. Slack connects through channels, threads and direct messages. Mindra can also coordinate agents a customer already runs.

SourceMindra, docs.mindra.co Connect apps and custom tools and Use Mindra in Slackread 2026-10-05

From a plain language goal Mindra assembles a team of specialized agents with a manager that plans the work, assigns roles and hands off across steps, agents can delegate to other agents, and it can also orchestrate agents a customer already runs. An agent is created with at least one connected capability, and a whole solution can be switched on with its agents and connections together.

SourceMindra, mindra.co/llms.txt and docs.mindra.co Create an agent and Enable a solutionread 2026-10-05

A workspace shared Documents library holds uploaded files in folders (up to 250 MB, videos up to 30 minutes) that agents search from chat, a persistent store separate from files attached to a single conversation.

Sourcedocs.mindra.co/docs/guides/add-documentsread 2026-09-27

Approval gates pause sensitive actions for a person to review in the app, in Slack or by email, the help center carries an approve actions guide, and policies and guardrails constrain what agents can do. Each connector action is set to Allow, Ask or Block.

In chat, a card asks for approval, and people can decide from a background panel while they keep working, approve a whole queue, or allow one request without changing the saved policy. Unanswered approvals expire, only an owner or admin can grant a one time organization override, and reversible actions and budget limits are enforced at run time.

SourceMindra, docs.mindra.co Approve actions and mindra.co/securityread 2026-10-05

Role based access control scopes people and agents, and SAML single sign on is supported. Mindra holds SOC 2 Type II, with independently audited security and availability controls, and is GDPR compliant with a DPA on request; a trust center sits at trust.mindra.co. Data is encrypted in transit and at rest, and secrets and credentials are handled on the server and never shown to agents. Tenants are isolated with least privilege service access, and prompts, data and outputs are never used to train a model.

SourceMindra, mindra.co/securityread 2026-10-05

Every agent action is recorded and exportable as a trail of who did what and when, every step is logged from the first prompt, and cost is tracked per agent. Each scheduled task keeps its own execution history, and chats are saved so they can be found again later.

SourceMindra, mindra.co/security and docs.mindra.co Schedule a task and Start a chatread 2026-10-05

Each agent team is described as having a shared memory, but Mindra does not publish what it holds, who can reach it, how long it lasts or how to delete it, and the help center has no memory guide. Mindra's marketing also speaks of compounding business memory and a record kept from the first prompt.

SourceMindra, mindra.co/llms.txt and docs.mindra.coread 2026-10-05

Mindra is a hosted service with no region choice, customer environment or self hosted option described. Zero Data Retention can be enforced end to end for sensitive work, and GDPR handling and tenant isolation govern the data, but none of them changes where the service runs. The web app runs at console.mindra.co.

SourceMindra, mindra.co/security and mindra.co/llms.txtread 2026-10-05

A Solutions catalog offers ready made sets of agents, connections and optional scheduled tasks filtered by industry, but no individual solution is named publicly; the use case pages describe workflows by team rather than products. Enabling a solution adds its agents, connections and scheduled tasks to the workspace in one step. Mindra describes ready AI departments for functions from sales, marketing and support to finance, HR, recruiting, RevOps and IT operations.

SourceMindra, docs.mindra.co Enable a solution and mindra.co/llms.txtread 2026-10-05

Tasks run on schedules at custom cadences, solutions can ship preset schedules, and the agents are described as proactive and running around the clock; people reach them from email, Slack or the web app. A task is created from any chat by saying what should happen and when, such as every weekday at 9 AM, and it starts only after a person approves the schedule card. Mindra also spots work on its own and suggests next steps.

SourceMindra, docs.mindra.co Schedule a task and mindra.co/llms.txtread 2026-10-05

Mindra routes each task to a suitable model or one the user chooses, across Claude, Gemini, GLM, Qwen, DeepSeek, MiniMax and others, and the help center covers choosing a model; premium models sit on the Pro plan and the free trial uses Gemini. The model picker locks models a plan does not include. A separate thinking setting, Standard or Enhanced, controls how deeply the model reasons without changing the model.

SourceMindra, docs.mindra.co Choose a model and mindra.co/llms.txtread 2026-10-05

The docs are a help center of nine task guides with no API reference, and no Mindra API, SDK, webhook or MCP server for calling the platform is published; MCP and REST appear only as tools Mindra calls out to. Earlier Mindra Console docs described a REST API with server sent events, and those docs are no longer available.

Sourcedocs.mindra.co/docsread 2026-09-27

Every step is logged and exportable for review after a run, and Mindra's blog discusses evaluating agents in production. Mindra publishes no way to test agents against a set of cases or to hold back a change that falls short.

SourceMindra, mindra.co/llms.txt and mindra.co/securityread 2026-10-05

Agents act through connectors, MCP and API tools; no browser or computer use capability is described. They work through OAuth connectors and registered REST or OpenAPI tools, and none of them drives a screen.

SourceMindra, mindra.co/llms.txt and docs.mindra.co Connect apps and custom toolsread 2026-10-05