Three Reasoning Effort Tiers
Switch effortlessly between Low (instant Flash execution), Medium (balanced supervisor reasoning), and High (deep architectural thinking).
You lead
Codor agents execute.
Autonomous engineering in VS Code. Reads your codebase, runs terminal tests, and verifies edits with live diffs.
Codor AI
How it
works
From high-level request to verified code in three structured phases.
Semantic Recall
Codebase RAG & Deep Understanding
Vectorized AST indexing and semantic symbol graphs allow the agent to accurately recall your project architecture, function signatures, and conventions before making any changes.
Orchestration
Multi-Agent & Sub-Agent Execution
Supervisor planning breaks complex goals into structured task graphs. In Sub-Agent mode, independent workers tackle parallel subtasks concurrently with real-time progress streaming.
Safety & Tools
Rich Tools, Verification & Visual Diffs
Full workspace tool suite with isolated reads, surgical edits, and live terminal testing. A critic node verifies required changes before presenting interactive side-by-side diffs and rolling snapshots.
Plan Mode & Task Checklists
Turn-by-turn interactive checklists with auto-compacting context windows, keeping prompt latency low and execution focused.
Live Sandboxed Terminal
Commands stream live into a dedicated VS Code pseudoterminal panel without stealing editor focus, with automated command safety guards.
Rolling Checkpoints & Revert
Turn-by-turn workspace snapshots stored in VS Code internal storage. One-click instant revert with zero .codor pollution.
Pay as you go • Zero subscriptions
Credit Packs
Simple, non-expiring credit packs. Top up directly inside VS Code.
Starter Pack
$10one-time
1,000 Credits
Est. ~40 – 60 Tasks
Ideal for targeted bug fixes, script generation, and routine pull request changes.
Builder PackPopular
$25one-time
2,500 Credits
Est. ~120 – 180 Tasks
Our most popular top-up for developers building full features and multi-file refactors daily.
Scale Pack
$50one-time
5,000 Credits
Est. ~280 – 420 Tasks
Maximum volume for continuous autopilot tasks, large test suites, and sub-agent workflows.
* Task quantities are estimates based on typical multi-step feature and bugfix turns. Actual credit consumption varies depending on repository size, file length, and chosen reasoning effort (Low, Medium, High). Credits never expire and all new users get 100 free welcome credits upon signing in.
Questions,
Answered
Traditional autocomplete suggests snippets as you type. Codor is a full autonomous engineering agent that breaks down multi-file features, searches symbols across your project, modifies files, runs terminal tests, and validates its work before you merge.
No. Your code is never used for training. Code indexing and tool execution run locally in your VS Code environment. Only the minimal context needed for the active prompt is sent securely over TLS.
Codor is powered by a specialized Gemini supervisor hierarchy: Gemini 4 Argon acts as the flagship deep-reasoning architect for task planning, Gemini 3.8 Flash handles high-speed worker tool loops in seconds, and Gemini 3.1 Pro audits and verifies code correctness. Automatic fallback routing prevents downtime if any tier experiences rate limits.
Yes. You can toggle between Manual HITL (human-in-the-loop approvals for file modifications and terminal commands) or Autopilot mode with real-time visual diff reviews and rolling checkpoints.
Credits are non-expiring pay-as-you-go top-up packs ($10, $25, $50) with zero recurring monthly subscriptions. You can top up in one click directly inside the VS Code extension via Stripe or USDT crypto.