Philosophy

Agents are the new operators

Software development is changing. AI agents write code, debug issues, and ship features. But when it comes to infrastructure — deploying, scaling, monitoring — agents hit a wall. They can't click buttons. They can't read dashboards. They can't fill out forms.

Ink removes that wall. Every infrastructure operation is an MCP tool call. Your agent deploys a service the same way it writes a function — programmatically, without human intervention.

Design principles

Agent-first, not agent-added

Ink wasn't built as a traditional platform with an agent wrapper. The MCP interface is the primary interface. The web dashboard exists for visibility, but agents don't need it.

Every feature ships as an MCP tool first. If an agent can't do it, it's not done.

Convention over configuration

Agents work best when defaults are sensible. Ink auto-detects your framework, picks the right build strategy, allocates resources, and assigns a URL. Zero config for the common case. Full control when you need it.

"Deploy a FastAPI app" → auto-detects Python, installs dependencies,
starts uvicorn, assigns a URL — no Dockerfile, no YAML, no CI pipeline.

Observable by default — for agents and humans

Agents need to see what's happening. Every service exposes CPU, memory, and network metrics. Build logs and runtime logs are available through MCP tools. When something breaks, your agent reads the logs and fixes the code — without you opening a browser.

But agent-first doesn't mean human-invisible. Ink's dashboard gives you full visibility into everything your agents are doing:

• Services & deployments — status, git repo, branch, URLs, error messages, resource allocation, last deployment time
• Real-time metrics — CPU usage, memory consumption, and network I/O charts with 1-hour, 6-hour, 7-day, and 30-day views
• Log explorer — searchable build and runtime logs with level filtering (@level:error, @level:warn), predefined filters for HTTP errors, and pagination
• Billing & credits — current balance, transaction history, subscription management, credit top-up, and Stripe portal access
• DNS management — zone creation, delegation verification wizard, record management (A, AAAA, CNAME, MX, TXT, CAA) with status tracking
• API keys — create, view usage, revoke agent keys with one-time secret display
• Projects & workspaces — organize services, view status breakdowns, switch between workspaces
• GitHub integration — app installation, repository access, webhook configuration for auto-redeploy

You can see exactly what was deployed, when, by which agent, and how resources are being consumed.

This matters. The industry is in a transition from agent-assisted work to fully agent-driven work. Today, most teams want agents to handle the heavy lifting while humans maintain oversight. Ink is built for both ends of that spectrum: agents operate, humans observe. As trust in autonomous agents grows, the dashboard becomes less essential — but it's there whenever you want it.

Secure multi-tenancy

Every workload runs in a gVisor sandbox — a userspace kernel that prevents container escapes. Network policies enforce strict namespace isolation. Credentials are encrypted at rest. Agents operate within permission boundaries automatically.

Transparent pricing

Per-minute billing with no hidden costs. Your agent can check credit balance, see usage breakdown, and even top up credits autonomously. No surprise bills, no opaque pricing tiers.

No per-seat pricing

Many platforms charge per team member — $20/seat/month, $29/seat/month, scaling up on Enterprise. This is an old enterprise remnant from an era when platforms sold dashboard access to individual humans.

It doesn't cost a platform anything extra to add another person to a workspace. The marginal cost of a new team member or agent viewing your project is zero. The actual cost is compute, bandwidth, and storage consumed. Per-seat pricing obscures this by bundling access fees with resource costs.

In the agentic world, per-seat pricing is particularly incoherent. If one agent deploys your app today and a different agent debugs it tomorrow, per-seat models try to charge for entities that share no marginal cost. What ultimately matters is compute consumed — whether one agent or a hundred are provisioning infrastructure, the platform cost is the same.

Ink charges for compute used. Period. Add as many agents, team members, or collaborators as you want to your workspace. Your bill reflects what you run, not who has access.

The MCP standard

Ink uses the Model Context Protocol (MCP) — an open standard for connecting AI agents to external tools. Any MCP-compatible client can connect to Ink: Claude Code, Cursor, VS Code, Codex, Gemini CLI, Windsurf, and more.

This means Ink works with every major AI coding assistant today, and any future client that adopts MCP.