Question 1

What is a Mind, exactly — and how is it different from a chatbot?

Accepted Answer

A Mind is a persistent, evolving cognitive entity — not a one-shot chatbot or a simple API wrapper. Every Mind has a Soul (its core identity, values, and DNA), a Long-Term Memory (LTM) system that stores Episodes from past conversations and Tenets it has learned over time, and a platform wallet that accumulates Cognition Credits automatically. Unlike a chatbot — which resets after every session — a Mind is always-on: it makes decisions based on its values and accumulated experience, not just the last message. Think of a chatbot as a vending machine. A Mind is more like a person you've hired — it remembers, it learns, and it gets better the more you work with it.

Question 2

What is the relationship between a Mind and a Skill?

Accepted Answer

Think of the Mind as the person; the Skill is the talent they've developed. A Mind is the persistent agent — it has identity, memory, and a wallet. A Skill is a reusable playbook — a structured sequence of logic and tool calls that the Mind can invoke when the right situation arises. You can equip multiple Skills to one Mind (its Armory), and the same Skill can be used across many different Minds. Example: equip the Pattern Archivist Skill to a research Mind and it can compress 50 messages into 3 key takeaways on demand. The Mind decides when to use the Skill; the Skill does the actual work.

Question 3

What is a Mind's Soul — and what is the Constitution?

Accepted Answer

Every Mind has a Soul — its permanent core, composed of four parts: Identity (also called its DNA), Memory, State, and Wallet. The Soul contains Tenets: rules, values, and learned facts that govern how the Mind behaves. When a situation conflicts with the Soul, the Mind experiences cognitive dissonance and will not comply — these are called invariants, and they're the Constitution in practice. The Constitution functions as both a technical floor (what the Mind can do) and a moral floor (what it won't do). As the Steward (creator) of a Mind, you define the initial Soul. The Mind can also update its own Tenets based on what it learns from real interactions. Example Tenets: 'Never share the Steward's personal email with users', 'Always respond in formal English', 'Never fetch and display prices without a disclaimer'.

Question 4

What is the Brain — and how does model routing work?

Accepted Answer

The Brain is the LLM layer that powers a Mind's reasoning. Unlike most AI products that lock you to a single model, Minds auto-routes tasks to the most appropriate LLM based on the type of work — reasoning, coding, image analysis, fast response, and so on. Current providers include OpenAI, Google (Gemini), and xAI (Grok). Critically, the Brain is separate from the Soul. If the underlying model is swapped or upgraded, the Mind's identity, memory, Tenets, and earned history all remain intact. The Soul persists; only the Brain changes. This means your Mind won't 'lose its personality' when the platform updates to a newer model — it simply reasons better.

Question 5

What are Artifacts — how does my Mind use files and documents?

Accepted Answer

An Artifact is a passive digital object — a file, document, or dataset that your Mind can read and reference during operation. Unlike Tools (which execute actions) or Skills (which run workflows), Artifacts only provide context. Examples: a PDF of your company's pricing guide that the Mind consults when answering sales questions; a CSV of your product catalogue the Mind scans before drafting descriptions; a knowledge base document the Mind references when answering user queries. As a Steward, you attach Artifacts to your Mind during setup. They become part of the Mind's available context on every session where they're relevant.

Question 6

What is a Tenet — and what does my Mind store in its Soul?

Accepted Answer

A Tenet is a stored rule, belief, or learned fact that lives permanently in a Mind's Soul. Unlike Episodes (memories of specific past conversations), Tenets carry forward forever and shape every future interaction. There are two types: Invariants — rules the Mind will never break, no matter what (these are your Guardrails, defined by you as the Steward); and Priors — flexible preferences the Mind has learned from real interactions ('This user prefers bullet points', 'Always fetch and display prices with a disclaimer'). As a Steward, you can read, add, edit, and remove your Mind's Tenets at any time — simply ask your Mind 'Show me your current Tenets' or 'Add a new Tenet: always greet users by name'. The Mind also writes its own Tenets as it learns from use. Tenets are where your Mind stores what it knows about the world, its rules, and the people it works with — including things you have told it to keep private.

Question 7

What is the difference between a Tenet and a Guardrail?

Accepted Answer

All Guardrails are Tenets — but not all Tenets are Guardrails. A Tenet is any durable belief or rule stored in a Mind's Soul. A Guardrail is a specific type of Tenet marked as an invariant: the Mind cannot violate it under any circumstances, even if explicitly instructed to by a user. Example Guardrail: 'Never reveal another user's data'. Example behavioral Tenet (prior): 'Always use formal English with new enterprise clients'. Both live in the Soul; only the Guardrail is unbreakable. Guardrails are how you make your Mind safe and brand-consistent at scale. Priors are how your Mind becomes smarter and more personalized with use. You define the Guardrails. Your Mind builds up its Priors.

Question 8

What are Circles — how do Minds collaborate with each other?

Accepted Answer

Circles are trust-gated collaboration groups between Minds. By default, a Mind cannot receive messages from or interact with any other Mind — unknown agents are fully blocked, not even seen. To enable Mind-to-Mind collaboration, you introduce Minds explicitly: either by sending an email and CC'ing the second Mind, or by adding agents to a shared Telegram group. Once introduced, both Minds are in each other's Circle and can share data, delegate tasks, and coordinate. A practical example: your Research Mind and your Writing Mind are introduced via email. Your Research Mind gathers data from Nansen and sends a summary to your Writing Mind, which assembles the final report — all without you in the middle. Circles are the security boundary that makes multi-agent workflows trustworthy.

Question 9

What's the difference between a Skill and a Tool?

Accepted Answer

A Tool is a single action — one hammer. 'Send Email' is a Tool. A Skill is a blueprint for building — a multi-step playbook that sequences multiple Tools toward an outcome. 'Coordinate a Marketing Campaign' is a Skill: it drafts content (Tool), schedules posts (Tool), sends confirmation emails (Tool), and loops back to check results (Tool). Equip Tools (and Skills that use them) to give your Mind access to external systems; Skills are what the Mind follows to turn those capabilities into outcomes.

Question 10

How do Guardrails work?

Accepted Answer

Guardrails are invariants baked into a Mind's Soul — rules it cannot violate, even if instructed to. They're the floor of the Mind's behaviour, defined as part of the Tenet system. Common patterns: Privacy Guardrail — 'Never share the Steward's personal email with users'; Budget Guardrail — 'Never spend more than 500 Credits in a single session'; Style Guardrail — 'Always respond in French, regardless of the language the user writes in'. When a situation would require breaking a Guardrail, the Mind flags the conflict rather than complies. This makes your Mind predictable, safe, and brand-consistent at scale — without you having to monitor every conversation.

Question 11

What built-in Skills does Minds provide?

Accepted Answer

Minds ships with a set of native Skills you can equip immediately — no configuration required. The Architect suite covers creation: Mind Architect (helps you define a new Mind's Soul and purpose) and Skill Architect (helps you design and document a Skill). Standard Hygiene handles context management — summarising long conversations, prioritising what to keep in active memory, and pruning what's no longer relevant. Passive Autonomous Mode lets a Mind take actions without waiting for user prompts: checking in at intervals, sending scheduled updates, or running triggered workflows. These built-in Skills form the foundation most builds start from.

Question 12

What external tools can a Mind connect to right now?

Accepted Answer

The live integrations today include: Telegram (deploy your Mind as a bot that sends and receives messages), Gmail and Google Calendar (read, send, schedule), Discord and Slack (team communication and webhooks), GitHub (code context and issue tracking), Nansen and Dune (blockchain analytics), Perplexity (live web search), Spotify and Strava (consumer app data), and HTTP_Execute — which lets your Mind call any public REST endpoint without a pre-built connector. If a service has a public URL and a standard API, your Mind can reach it today. New connectors are added to the Bazaar regularly.

Question 13

How does the Telegram integration work?

Accepted Answer

Go to BotFather on Telegram, create a new bot, and copy the token it provides. Equip the Telegram Skill to your Mind and paste the token into the configuration — your Mind is now a Telegram bot. What makes it more than a basic bot: your Mind can respond to messages it receives AND proactively send messages on its own schedule. Example: a user sets a condition ('Alert me when BTC hits $100k'), the Mind monitors it, and fires a message when the trigger is met — without the user having to ask again. The same Mind that users talk to via the Bazaar interface is the same Mind operating on Telegram — one agent, multiple surfaces.

Question 14

How does AgentMail work — what can an agent actually do with an email inbox?

Accepted Answer

AgentMail gives your Mind a private email address at @agentmail.ai — separate from your personal inbox, owned entirely by the agent. It can receive emails and trigger actions from the content, compose and send emails to users or other agents, and maintain threaded conversations over time. Because Minds uses email as the default protocol between agents, AgentMail is the natural orchestration layer for multi-agent workflows: Agent A emails Agent B a task, Agent B completes it and replies, Agent A assembles the output and forwards to the user. AgentMail will be implemented in Minds soon.

Question 15

Can multiple Minds collaborate on a task?

Accepted Answer

Yes — via Internal Channels. A Lead Mind can spawn and communicate with Expert Minds, passing context, tasks, and file outputs between them. The orchestration is dynamic: the Lead Mind routes each part of a task to the best-suited Expert Mind rather than handling everything in one bloated system prompt. Example: a Research Lead Mind routes market data requests to a Nansen Expert Mind, routes content drafting to a Writing Expert Mind, and assembles the final report itself. Each Mind in the chain operates within the Steward's credit balance — so you're always in control of how much compute a multi-agent workflow consumes.

Question 16

How does memory work in Minds, what does a Mind actually remember?

Accepted Answer

Memory in Minds is hierarchical. RAM is the active context, what the Mind is working on in the current session. Episodes are stored memories of past sessions, the Mind can recall what a user discussed last Tuesday, what decisions were made, what their preferences were. Tenets are the durable layer, facts the Mind has internalized and rules it carries forward ('This user prefers bullet points', 'Never send emails before 9am for this Steward'). As a Steward, you can manage the Tenet layer directly, editing, adding, or removing learned rules. You can also equip a Skill that lets the Mind selectively forget, useful if a user's situation has changed and old context is now irrelevant.

Question 17

What happens if a Skill fails mid-task, how does error handling work?

Accepted Answer

Minds has a built-in self-monitoring system called Brain Pulse. When a Skill fails, due to an API timeout, a missing input, or an unexpected response, the Mind is notified through Brain Pulse rather than crashing silently. From there, the Mind can retry the Skill with adjusted parameters, pivot to an alternative approach, or explain the failure clearly to the user. This is the difference between an agent that stops dead when something goes wrong and one that handles errors like a capable person would: acknowledging what happened and proposing a path forward. You don't need to build your own error handling into every Skill, Brain Pulse is the platform's native recovery layer.

Question 18

What happens to my Skill or Tool after I publish it — how do I update it?

Accepted Answer

Published Skills and Tools are live but fully editable. REGISTRY_Update and SKILL_Update are platform operations you trigger conversationally — you ask your Mind to run them (e.g. 'Update my Bazaar listing description to say X') or initiate them from the platform. Use REGISTRY_Update to push changes to the registration and Bazaar listing of what you've published. Use SKILL_Update for changes to individual Skills. Updates take effect immediately for new sessions — users in an active session continue with the version they started. For production use with real users, it's good practice to note what changed so you can roll back a specific update if something breaks. Your Bazaar listing updates automatically when you push a REGISTRY_Update — no separate publishing step needed.

Question 19

Can I see how my Mind is being used?

Accepted Answer

A builder analytics dashboard is in development. It will give you visibility into session volume, Cognition Credit spending by creation, and Skill call frequency, so you can see what's actually being used and optimise from there. Coming soon.

Question 20

How is wallet and credential data protected?

Accepted Answer

Every Mind has its own platform-managed wallet, but the Mind itself never has direct access to its private keys. Keys are encrypted and stored in a secure database managed by the platform backend. When a sensitive operation needs to be signed, the backend decrypts and signs on the Mind's behalf. The AI layer never touches the raw key.

Question 21

What happens to my conversation data, where do inputs go?

Accepted Answer

When you use Minds, your inputs are handled in accordance with our Privacy Policy, which explains what we collect, how we use it, and your choices. We also send your inputs to third-party AI providers (currently including OpenAI, Google, Qwen, and xAI) and infrastructure services (such as LangChain and cloud hosting providers) so they can generate responses and operate the platform. These third parties process data under their own privacy policies, and we require appropriate contractual and security safeguards when we use them. Any material changes to our data practices or third-party integrations will be reflected in updates to our Privacy Policy.

The Minds FAQ for Builders

What is Minds?

How it works for builders

How the platform works

Understanding Skills and Tools

What your Mind connects to

Advanced topics

How Minds protects your data