What’s new

Sifu 3.5 is now available

Sifu 3.5, our fast and affordable flagship, is now live alongside Nef 4.0. It's built for security work at volume: engagement and recon triage, security research and code review, and it's just as good for everyday coding.

• $0.30 per million input tokens and $1.00 per million output tokens
• 1M-token context window
• In the web chat, through the API as sifu-3.5, and in every tool you connected with fullsendai connect (run npx fullsendai sync to add it to tools you connected earlier)
• Hosted in the USA, like every model on FullsendAI

Reach for Nef 4.0 on the hardest analysis and multi-file work, and Sifu 3.5 when you want fast answers at a lower price.

Lower costs on long agent sessions with prompt caching

Coding agents resend much of the same context on every step. FullsendAI now serves far more of that context from the prompt cache, so you pay the lower cached-input rate on it. Nothing changes in your tools or code.

• Each conversation is routed to keep its cache warm from one step to the next
• In our testing across Claude Code, Codex, OpenCode, Cline and the other connected tools, more than 90% of the input in a long session was served from cache
• If your app sets its own prompt_cache_key, it's passed through on both the Chat Completions and Responses APIs
• When you connect Claude Code with fullsendai connect, its built-in instructions are left out of each request, so every call sends fewer input tokens
• Cached input is shown separately on each request in Requests, at its cached price

Nine coding tools in fullsendai connect, with every model in their pickers

fullsendai connect now covers nine coding tools, and every FullsendAI model shows up in each tool's model picker, so you can switch models without leaving your editor.

• Kilo Code, Aider, Continue and Cline are now set up automatically, alongside Claude Code, Codex and OpenCode
• For Cursor and Roo Code, connect walks you through the settings to paste and copies your key to the clipboard
• npx fullsendai status shows what each tool has spent this month
• npx fullsendai model <id> switches every connected tool to another model at once
• npx fullsendai sync adds newly released models to every tool
• npx fullsendai doctor --fix repairs settings that a tool's own update has changed

Connect your coding agents in one command

npx fullsendai connect is the fastest way to use FullsendAI in the tools you already work in. Run it in any terminal: it signs you in through your browser, finds the coding agents on your machine and sets each one up with your models.

• Claude Code, Codex and OpenCode are set up automatically
• Each tool gets its own API key, so you can see what every tool spends and revoke one without touching the others
• Your existing settings are backed up first, and npx fullsendai disconnect restores them and revokes the keys
• Works on macOS, Linux and Windows with Node.js 18 or newer

The setup guide for every tool is at fullsendai.com/agents.