Introducing CodeGrex: The AI-Native IDE Built for How You Actually Code

CodeGrex completions go far beyond simple token prediction. The engine indexes your open files, imports, and recent edits to produce multi-line suggestions that are syntactically correct and contextually relevant — often finishing an entire function body in one Tab press.

Toggle Supercomplete for larger, speculative completions. It pre-fetches multiple candidate continuations and ranks them by probability, letting you cycle through options with a single keystroke. Perfect for boilerplate-heavy code.

CodeGrex's routing engine picks the optimal model for each request in real time. Small completions go to a fast local model for sub-100ms latency; complex multi-file refactors route to a more powerful cloud model. You set the policy — the router handles the rest.

Describe a task in plain English — "add pagination to the users API" — and the Agent creates a step-by-step plan, edits multiple files, runs your tests, and shows you a clean diff for review. It can invoke terminal commands, read logs, and retry on failure, all within guardrails you control.

Need a task done but don't want to block your current work? Background Agents spin up an isolated git worktree, execute the plan on a separate branch, and present a PR-ready diff when they finish — all while you keep coding on main.

For large-scale refactors, CodeGrex can coordinate multiple agents working in parallel across different parts of the codebase. A supervisor agent manages task distribution, conflict detection, and merge resolution automatically.

Agents can execute shell commands, run test suites, inspect build output, and react to errors — all with explicit user permission per session. Every command is logged in an auditable history.

CodeGrex ships with first-class Ollama integration. Browse, pull, and manage local models directly from the IDE. When a local model is running, completions and chat work entirely offline — no data ever leaves your machine.

Choose from three privacy levels: Normal (cloud + local), Sensitive (local-only completions, cloud chat), or Offline (fully air-gapped). The setting applies globally or per-workspace for mixed-sensitivity projects.

CodeGrex collects no usage telemetry unless you opt in. Code snippets, prompts, and completions are never stored on our servers unless you explicitly enable cloud analytics for your organization.

CodeGrex maintains a local vector index of your workspace. When you ask a question or trigger a completion, the engine retrieves the most relevant files, functions, and documentation — not just what's open in your tabs.

Set context depth from minimal (open files only) to full (entire repo). CodeGrex adjusts token budgets automatically so you get the richest context that fits within model limits.

Connect external tools — databases, APIs, documentation servers — through the Model Context Protocol. Agents can query your Postgres schema, fetch Jira tickets, or read Confluence pages to ground their output in real project context.

Use your own OpenAI, Anthropic, or Google API keys. Requests route directly to the provider — CodeGrex never proxies or stores your key. Perfect for teams with existing billing or compliance requirements.

Admins can enforce model allow-lists, token limits, tool permissions, and privacy modes across the entire org. Policies propagate automatically — developers get the right defaults without manual configuration.

Every agent session, prompt, and tool invocation is recorded locally. Enterprise plans add centralized audit logs for compliance teams, with configurable retention and export options.

CodeGrex is the real thing, not a wrapper. Your extensions, themes, keybindings, and settings carry over. If it works in your current editor, it works in CodeGrex — there's nothing new to learn.

Trigger AI actions directly in your code: explain a function, generate tests, refactor a block, translate to another language. Custom slash commands let you define your own AI-powered workflows.

Chain multiple AI steps into reusable Flows. For example: "read the error log → identify the root cause → generate a fix → run tests → open a PR." Flows are shareable across your team.

Change your agent, routing mode, privacy level, and model preferences from a sleek bottom-bar popup — no digging through JSON. Settings apply instantly and sync across sessions.