
Kilo
Code with AI without breaking the bank
About Kilo
Kilo is an agentic coding platform whose actual argument is about margin. Every subscription coding assistant makes money on the spread between what inference costs and what you pay, and Kilo attacks that directly: the platform is MIT licensed and free, and its gateway routes to more than 500 models at exact provider rates with zero markup, alongside bring your own key and fully local model support. If you do not trust the gateway, you can point it at your own key or run models locally and the product still works.
The surface area is unusually wide for a project first published in March 2025. Kilo runs as a native VS Code extension, a native JetBrains plugin covering IntelliJ, PyCharm and WebStorm, a command line interface, a browser extension and a hosted cloud, with sessions that transfer between platforms and fork into branches. Work is organised into modes rather than one undifferentiated chat: Code, Architect, Debug and any custom mode you define, so the tool behaves differently when you are planning than when you are fixing. The project was acquired by Anaconda on 2026-07-15 and continues to ship as open source, which puts it in front of a very large existing Python and data science install base.
Install the extension for your editor, or use the CLI if you live in a terminal. Pick a mode for the task at hand: Architect when you want a plan before any code is written, Code for implementation, Debug when you are chasing a fault, or a custom mode you have defined for your own workflow. Choose how inference is paid for, which is the decision that distinguishes this tool. Auto Free, bring your own key and local models all cost nothing through Kilo; the Kilo Gateway charges provider rates with no markup and no subscription; Kilo Pass buys credits in bulk with bonus credits attached. Auto Model can make the per task choice for you, weighing the work against the budget. As you work, codebase indexing enables semantic search and context mentions such as git changes, folder and terminal pull the right material in without manual pasting. A memory bank carries project context across sessions, and cloud agents run tasks off your machine, with an agent manager for running several in parallel.
- •Zero Markup Gateway - Routes to 500+ frontier, open weight and custom models charged at exact provider rates, alongside BYOK and fully local models
- •Agent Modes - Code, Architect, Debug and user defined custom modes, so behaviour matches the task
- •Auto Model - Selects a model per task based on the work and the budget rather than a fixed default
- •Native Editor Support - VS Code and JetBrains plugins, a CLI, a browser extension and a hosted cloud
- •Cloud Agents - Run tasks without consuming local resources, with an agent manager for parallel work
- •Automated Review Agents - Pull request analysis plus a separate security agent for vulnerability review
- •Memory Bank - Retains project context across sessions, with sessions transferable between platforms and forkable into branches
- •Codebase Indexing - Semantic search across the repository, with context mentions for git changes, folders and the terminal
- •Team Controls - Centralised bring your own key management and an AI adoption score on team accounts
Developers and teams paying for a coding assistant who have looked at the token bill and wondered where the money goes. The no markup gateway plus BYOK plus local models gives cost sensitive teams a concrete reason to switch rather than a positioning claim, and centralised key management makes that workable across a team rather than per developer. The breadth of editor support means a team using both VS Code and JetBrains does not have to standardise first. Python and data science teams already inside the Anaconda ecosystem have the shortest path in. Two honest caveats: the Anaconda acquisition is very recent and the no markup promise is the easiest thing for an acquirer to walk back, and this is crowded territory competing directly against Cline, Roo Code, Continue and every vendor native agent.
Pricing
$0 - $15/mo
- Free & Open Source$0/mo
- Teams$15/mo
- EnterpriseContact sales
From the vendor pricing page, 2026-09-16














