The AI agent platform, in full

Every capability, for the person doing the due diligence.

The home page makes the argument. This page has the detail behind it: every capability that ships with Crios, everything it connects to, and the questions teams ask before they deploy.

Back to the control room

Everything included

Every AI agent capability, in the box

Every capability below ships with Crios. Open a group to see what is inside.

Planner → Coder → Reviewer → Supervisor

Every session runs a staged pipeline of specialized agents that plan, implement, critique, and gate each other’s work.

Plain-English briefs

Start from a freeform description, a GitHub issue, a Jira ticket, or a Slack command. Crios turns it into a concrete build plan.

Verified before shipped

Tests, security review, migration lint and a vision-model gate on the rendered app. Real pixels, not self-reported success.

Live steering

Pause, resume, retry, or send follow-up guidance mid-build. You stay in control without babysitting.

Web, mobile and desktop targets

About 75 templates, Android emulator and desktop sandboxes, and design reconstruction from a screenshot.

Parallel multi-agent execution

Coordinator/worker fan-out with configurable concurrency budgets for large builds.

Stored, versioned agents

Persona or raw instructions, model and fallback, tool allowlist, delegation, knowledge filters, structured output.

Guardrails

Input and output checks as expressions, LLM rubrics or workflows, with retry-on-fail.

Tool approvals and budget gates

Approve, edit or reject tool calls; max iterations and cost with park-for-approval.

Durable memory

A Memory node on the project database, browsable and editable, with a separate test partition.

Runs, watch and rehearse

Per-version success rate, latency and cost; watch child steps on the canvas; rehearse without side effects.

Skills and Skill Studio

Portable, permission-scoped capability packages drafted from a description and run in isolated pods.

About 370 node types

AI, logic, data, platform actions and 255 integration nodes on a visual canvas.

80 triggers

Schedules, webhooks, forms, files, queues (Kafka, RabbitMQ, MQTT, Redis, SNS), feeds, email and 60 vendor events.

Prompt to workflow

Describe it; the AI drafts it with clarifying questions and leaves blanks where it lacks a value.

Durable executions

Retries, timeouts, circuit breakers, dead-letter queue with redrive, versions with diff and restore.

External intake

GitHub issues, Jira tickets and Slack commands become sessions, with an org inbox and write-back.

Collectors and signals

Feeds and APIs classified into named signals that trigger workflows; an AI scout finds working sources.

Project discussion channel

Human-first, Slack-like, with message kinds, topics, mentions, attachments and moderation.

Decisions tracker

Candidate, confirmed, dismissed, superseded and reversed decisions, promoted into the knowledge base.

crios_bot

Classifies messages, proposes decisions, recalls prior rulings and answers when summoned or certain.

Tasks from the thread

File @todo items into a lightweight project task board linked to messages and decisions.

Conversations become memory

Messages are embedded once immutable, so past discussions surface in future work.

Expert routing

Topics route to the right stored agent; ask agents to weigh in with one reply per message.

One-click publish

Deploy finished projects to your cluster behind your own domain with automatic TLS certificates.

Managed project databases

Postgres, MySQL, Mongo- and Redis-compatible engines or BYO; sandbox and durable tiers; one-click export.

Publication monitoring

Synthetic probes, traffic, uptime, latency, logs and alerts for every published app.

Live preview environments

Inspect work-in-progress apps on preview URLs while the agents are still building.

Git-native output

Every build lands as commits and pull requests on GitHub, GitLab, or the bundled Gitea.

LicenseWatch

Open-source license compliance for dependencies, with policy, findings and acknowledgements.

Multi-provider by design

Anthropic, OpenAI, Google Gemini, OpenRouter, Fireworks, Kimi and MiniMax, plus any compatible or local endpoint.

Per-stage model routing

Plan on one model, code on another, review on a third. Configurable per org, project and session type.

Fallback chains

Automatic failover to backup providers on errors or outages, mid-session, without losing work.

Bring your own keys

OpenAI- and Anthropic-compatible custom endpoints, including self-hosted and local inference.

Live pricing sync

Provider price catalogs refresh continuously, powering cost-aware routing decisions.

Budgets & cost consoles

Org budgets, quotas, per-session forecasts, burn-rate telemetry and CSV export.

Org knowledge base

Vector-embedded docs, patterns, and domain facts searchable by every agent during builds.

Project-scoped memory

Per-project conventions, architecture decisions, and standards that agents consult automatically.

RAG-injected context

Relevant knowledge is retrieved and injected during planning and coding, with no manual prompting.

Failure → fix learning

A human-reviewed suggestion queue turns past failures and their fixes into playbook rules.

Markdown KB editor

Create and edit knowledge entries in the UI with org/project scoping and full history.

Built-in web search

Bundled SearXNG gives agents private, self-hosted web search during builds.

Ephemeral Kubernetes sandboxes

Every session gets an isolated, disposable workspace pod with managed lifecycle and resource budgets.

Kata Containers microVMs

Per-pod lightweight VMs make the hypervisor the trust boundary, not the container runtime.

Locked-down runtime

Restrictive egress network policies, non-root execution, read-only root filesystems, dropped capabilities.

Vault-managed secrets

Provider keys and integration credentials live in HashiCorp Vault, injected per session, never logged.

SSO, MFA and RBAC

OIDC single sign-on, TOTP with backup codes, roles, groups, invitations and scoped personal access tokens.

Audit everything

Comprehensive action logging with webhook export for your SIEM and compliance workflows.

Single Helm chart

The whole platform, from orchestrator and UI to sandboxes and data stores, deploys with one helm install.

Any Kubernetes

k3s and OrbStack for dev, EKS/GKE in the cloud, Talos Linux on bare metal for production.

Air-gap friendly

Everything runs in-cluster: Postgres, Redis, Qdrant, Gitea, Vault, web search. The only egress you need is to your model providers, and those can be local too.

Observability included

Prometheus metrics, Grafana dashboards, and Loki log aggregation ship in the chart.

Backups and disaster recovery

Scheduled backups to S3-compatible storage with test and run-now, restore runbooks and retention.

10 languages

English, Chinese, Hindi, German, French, Japanese, Spanish, Portuguese, Turkish and Korean across the whole UI.

Integrations

Works with the tools your company already runs

Connected accounts for your agents, hundreds of workflow nodes and connector types, and Crios itself as a tool in your AI editor.

Social & publishing

X · LinkedIn · Reddit · Instagram · TikTok · YouTube · Facebook Pages · Facebook Lead Ads

Git & CI

GitHub · GitLab · Gitea · Bitbucket · Jenkins · CircleCI · Travis CI · Sentry

Ticketing & projects

Jira · Linear · Asana · ClickUp · Trello · Monday · Notion · Todoist

Chat & messaging

Slack · Discord · Microsoft Teams · Google Chat · Telegram · WhatsApp · Matrix · Mattermost

Email & SMS

Gmail · Outlook · SMTP / IMAP · SendGrid · Mailgun · Mailchimp · Brevo · Customer.io

CRM & marketing

Salesforce · HubSpot · Pipedrive · Zoho · Dynamics 365 · Copper · Segment · PostHog

Files & documents

Google Drive · Docs · Sheets · Slides · OneDrive · SharePoint · Excel · Dropbox

Databases & data

PostgreSQL · MySQL · SQL Server · MongoDB · Redis · Supabase · Snowflake · Databricks

Cloud & identity

AWS · Cloudflare · Azure Monitor · Okta · LDAP · Bitwarden · Graph Security

Security & ops

CrowdStrike · Elastic Security · MISP · TheHive · Cortex · Splunk · PagerDuty · UptimeRobot

Commerce & finance

Stripe · PayPal · Paddle · Chargebee · QuickBooks · Xero · Shopify · WooCommerce

Connected accounts

OAuth accounts your agents act through, with least-privilege scopes and owner approval on sensitive actions.

Gmail · Google Calendar · Google Drive · Slack · Discord · Jira · Facebook Pages · Instagram · TikTok · X

More than 250 integrations, about 400 node types, 80 triggers and 300+ connector credential types. Anything else reaches through the HTTP node.

Connect your AI

Drive Crios from Claude Code, Cursor or Claude Desktop

Crios is an MCP server: start builds, steer sessions, publish, search knowledge and run workflows from the editor you already use.

Authenticated with a personal access token you can revoke at any time.

claude mcp add --transport http crios https://mcp.crios.ai/mcp \ --header "Authorization: Bearer crios_pat_…"
FAQ

Questions teams ask before deploying AI agents

Can Crios run air-gapped?

Yes. Postgres, Redis, vector search, git hosting, Vault and web search all run inside the cluster. The only outbound traffic you need is to your model providers, and those can be local or self-hosted endpoints.

Which models can I bring?

Anthropic, OpenAI, Google Gemini, OpenRouter, Fireworks, Kimi and MiniMax are built in, plus any OpenAI- or Anthropic-compatible endpoint, including self-hosted inference. Each stage of work can route to a different model with fallback chains.

Where does captured knowledge live?

In your deployment. Knowledge entries, project memory, decisions and discussion history are embedded into a vector store inside your cluster and never shared with another tenant or with us.

What is the difference between an agent and a workflow?

A workflow is a deterministic graph of nodes with triggers, logic and integrations. An agent is a stored, versioned model-driven worker with tools, memory and gates. Agents run inside workflows as nodes, and workflows can be tools of agents.

Can non-engineers use it?

Yes. Work can be filed from Jira, GitHub or Slack, workflows and agents can be drafted from a plain-language description, and the Discussion channel is a human conversation where agents chime in when useful.

How are approvals enforced?

Gates are part of the definition and enforced by the platform: a tool call marked for approval parks the run until an assignee decides on the web or in Telegram, optionally editing the arguments. Budget limits park or fail the run. Everything is audit-logged.

Is Crios Cloud a shared SaaS?

No. Crios Cloud provisions a dedicated cluster for your organization on bare-metal servers you choose, billed monthly. Nothing is shared with other tenants.

How do I connect Claude Code or Cursor?

Crios is an MCP server. Add it to your editor with a personal access token and you can create projects, start builds, steer sessions, publish and run workflows without leaving your tools.