Install the free MIT agent kernel from npm, run it on your own machine or server, and give agents real tools, persistent memory, channels, and optional skill packs.
23 live providers, BYOK, local backends, and automatic fallback chains.
19 channel adapters: CLI, web, Telegram, Slack, email, API, and more.
Optional skill packs; install only what you need.
Use the free kernel directly today. Add domain packs as you need them, and upgrade to Scout only when a production agent needs governance.
Run Carina locally or on your own server with Node.js 22+. Start with CLI chat, then add providers, memory, and channels.
Install only the domain knowledge your agent needs. Packs keep Carina general-purpose while letting teams specialize workflows.
Use the same kernel behind a product, internal workflow, or hosted service. Carina gives you the agent runtime so you can focus on the job.
When an agent touches sensitive production work, add Scout for monitoring, approvals, audit exports, and kill switch controls.
UK property investment platform running on Carina with deal analysis, HMO checks, and property workflow support.
Fleet tracking for Africa. Logistics agents on Carina with regional channel adapters.
NHS compliance copilot for policy classification and audit-ready agent workflows.
Borehole industry platform for Ghana and West Africa. First Africa distro on Carina.
The kernel plans steps, calls tools, and reports back. Free to self-host; MIT licensed. Add Scout when the same agent needs production monitoring, approvals, and incident response.
Research, file edits, shell commands, browser steps, email. Carina acts on your stack instead of handing you a paragraph to copy.
Past conversations and outcomes stay available across sessions so repeat work does not start from zero.
CLI, web UI, Telegram, WhatsApp, Slack, or API. Same agent, same memory; you choose the channel.
Carina Core is general-purpose. These pages show who each setup is for, what to configure, and one example workflow.
UK deals, HMO checks, property-uk pack and RAG.
Triage tickets, draft replies, save files for review.
Web and arXiv briefs with research-basics.
Run Carina in the browser, on your own server, or both.
legal-uk and finance-uk skill packs on Carina; EU AI Act classification, approval workflows, and board reports on Scout.
Hub page plus links to 16 optional skill packs.
Self-host with Node.js 22+ on Linux, macOS, or Windows. Add Postgres, Redis, and Docker Desktop when you need persistent memory, queues, and full shell or Python sandboxing. Use Carina Cloud if you want a hosted workspace.
Use Node.js 22 and npm today. PowerShell direct shell mode works; Docker Desktop is required for stronger shell isolation and Python code execution.
Run the CLI and web gateway on your own VPS with Postgres, Redis, and optional nginx when you need a persistent workspace.
Pair with Scout for prompt guard, approvals, live events, audit exports, and a kill switch before agents touch sensitive workflows.
Bring your own keys, govern your agents, audit everything, control spend. Carina Ops is the paid operations console, coming soon on app.carinaai.uk.
Connect OpenAI, Anthropic, DeepSeek, and other providers with encrypted keys at rest.
Run test prompts against your keys before you wire agents into production workflows.
Org-level token spend, credits, retention by plan, and exportable audit trails on Team.
Free agent for personal use and small projects. Carina Ops is coming soon for teams that need BYOK, audit, and spend controls.
Need agent governance and protection? See Labyrinth Scout at labyrinthscout.com
Domain packs and products build on the MIT kernel. Scout is the optional commercial security layer.
Verified counts as of 12 June 2026. Shipped vs configured vs optional: canonical capability inventory.
Base registry; toolsets enable subsets. Web search, files, shell, HTTP, email, code, browser, MCP, and more.
Catalog entries; chat uses configured keys or local backends. Switch mid-session with /provider.
CLI, web UI, Telegram, WhatsApp, Slack, Discord, email, API server, and more when configured.
96 curated skills across engineering, finance, legal, devops, property, and more.
Carina can draft new skills after solving complex problems when auto-writing is enabled.
Vitest coverage across the agent core. See the public repo for CI status.
| Carina capability | Included in Core |
|---|---|
| Autonomous task execution | |
| Persistent memory with semantic search | |
| Optional skill writing | opt-in |
| 16 skill packs (96 curated skills) | optional |
| 19 channel adapters | when configured |
| 66+ tool execution surface | toolset-dependent |
| Switch LLM mid-session (51 catalog; configured subset) | when configured |
| npm package (carina-agent) | v0.2.0 |
| MIT-licensed source | GitHub |
Agent governance (EU AI Act, audit trails, approvals, industry templates) lives in Labyrinth Scout. Capability counts reflect shipped definitions; availability depends on configuration. See the canonical capability inventory and labyrinthscout.com.