The editor, the agent and every model you connect are free with your own API keys. Pro is optional: it turns on the hosted services, embeddings, web search, MCP and voice, for $5 a month per seat, half the usual price. Comparing editors? Read Cortex vs Cursor.
All AI models use your own API keys. You pay those providers directly. Pro is priced per seat, the standard subscription model for AI tools: one developer, one seat. The seat charge is a platform charge for the hosted services behind web search and voice assistance, never a markup on your tokens.
There is no free trial to run out and no feature gate on the agent itself. The free tier is the whole IDE; Pro only switches on the services that have to execute on a server we operate. Everything in the first column keeps working whether or not you ever subscribe, and it keeps working if you cancel.
| Feature | Free (BYOK) | Pro Monthly $5/seat/mo |
Pro Yearly $40/seat/yr |
|---|---|---|---|
| Agentic loop: plan, edit, run, verify | Included | Included | Included |
| Full codebase context and indexing | Included | Included | Included |
| Terminal, Git and file tools | Included | Included | Included |
| Any model, any OpenAI-compatible provider | Included | Included | Included |
| Hosted semantic search (embeddings) | Local fallback | Included | Included |
| Web search and fetch for the agent | Not included | Included | Included |
| MCP server connections | Not included | Included | Included |
| Voice input (speech to text) | Not included | Included | Included |
| Support | Community | Community | Priority, 24h |
| Model usage | Always billed by your provider at list price. Cortex never meters, marks up or resells your tokens. | ||
Each Pro feature exists because it needs infrastructure we run: an embedding model, a search index, a speech-to-text service, or an MCP relay. The subscription is a per-seat platform charge for exactly that, one seat per developer, which is how hosted AI capability is priced across the industry. It is why cancelling never breaks the editor.
Your project is embedded so the agent can find code by meaning, not only by exact text. Ask "where do we throttle API calls?" and it lands on the rate limiter even though that word never appears. Without Pro, search falls back to a fully local method, so nothing leaves your machine but recall is lower.
The agent can look up current documentation, changelogs and error messages mid-task instead of guessing from training data. Results are retrieved on our servers and cited back into the conversation.
Model Context Protocol servers attach external tools and data, issue trackers, databases, browsers, internal APIs, to the agent as first-class tools. See the MCP setup guide and the 25 servers worth connecting.
Dictate a prompt or a code review note and it is transcribed into the chat, so long instructions do not have to be typed.
The editor, the agent loop, all 13 core tools, terminal and Git access, multi-file edits, memory, permissions, local model support through Ollama or LM Studio, and every provider you add through Cortex Connect. None of it is time-limited.
Pro Yearly is the same feature set as Pro Monthly for $40 instead of $60, and it adds priority support with a 24-hour target plus early access to features before they reach the monthly plan. If you are still weighing the model, read what BYOK means and what agent work actually costs per provider.
Yes. The IDE is free when you use your own API keys (BYOK). You pay your AI provider directly, at that provider's list price per token. Cortex adds no markup, no credit meter and no usage cap of its own.
Pro adds the services that run on our servers: embeddings for semantic code search, web search and fetch for the agent, MCP server connections, and voice input. It is a per-seat platform charge of $5 per month or $40 per year, and your model tokens stay billed by your own provider at list price.
No. With BYOK, prompts and code go straight from your machine to your AI provider. API keys are stored locally in your operating system's secure credential store, and your project never leaves your disk.
Cortex ships native builds for Windows 10, Windows 11 and Linux, with no WSL or container required. There is no prebuilt macOS installer yet; macOS is available via the official maintainer-authorized build guide, which walks you through building from source in about 30 to 45 minutes.
The IDE keeps working. You lose the hosted services at the end of the billing period, semantic search drops back to its local method, and web search, MCP and voice turn off. Your keys, projects, memory and settings are untouched, because they were never on our servers.
No. Token usage is billed by your provider, at list price, on whatever plan you already have with them. That is why heavy agent use usually costs less here than in an editor that resells inference: routing cheap models to cheap tasks lowers the bill immediately, with no margin in between.
Ready to try it? Download Cortex free, sign in with a free account, add one API key, and open a project. The setup takes about five minutes and the docs cover every step. Comparing editors first? See Cortex vs Cursor: BYOK pricing compared for the month-by-month cost difference at real usage levels.