FloatBoat Blog

GPT-6 Astra — OpenAI's Most Capable, Most Guarded Model

GPT-6 Astra is OpenAI's most capable and most restricted model: specs, pricing, the ARC-AGI-3 dispute, Critical cyber status, and what solo builders should do.

Finding Raycast for Windows? Try Floatboat!

Searching for a Raycast alternative on Windows? Discover how Floatboat goes beyond a launcher with an AI-native workspace, Combo Skills, local context, and autonomous workflows.

Gemini 3.8 Flash — Google's New Coding & Agent Workhorse

Gemini 3.8 Flash is Google's new coding-and-agent workhorse: DeepSWE 73.7%, a 1M context, and a gated Cyber twin. How to use it in Floatboat with zero setup.

MiniMax H3 Max Live Stream — Faster Than Playback

MiniMax H3 Max live stream on fal: generate video faster than playback for infinite AI broadcasts. H3 Max Live, Twitch demos, costs, and architecture.

Codex Harness Open Source — OpenAI's Agent Runtime

OpenAI's Codex harness is open on GitHub: exec, SDK, and app-server. What's in the agent loop, what's not open, and how it compares to DeepSeek Harness.

豆包工作是什么 — 字节跳动独立 AI 办公 Agent 完整定义、本地操控与飞书集成能力说明

豆包工作 Doubao Work 是字节跳动 2026 年 8 月正式发布的 AI 办公 Agent 平台，支持 Windows 虚拟桌面操控、飞书企业深度集成与云电脑后台任务。本文解析产品定义、为何独立成品牌，以及与 WorkBuddy、千问办公、Floatboat 的竞争格局。

Introducing Floatboat Flow Mode — AI Voice Dictation

Floatboat Flow Mode turns speech into living documents—edit while you dictate, batch voice annotations for your Agent, and get meeting action items before the call ends.

How to Automate Weekly TikTok Shop Research with FastMoss and Floatboat

See how Floatboat operates FastMoss in its built-in browser, saves live evidence, compares history, writes and checks a management brief, delivers it, and returns through FloatSchedule.

GLM-5.3 — Post-Training Scaling, Open-Weight Coding, and the Security Surprise

GLM-5.3 keeps GLM-5.2's base and gains ~50% coding through post-training scaling. Terminal-Bench, DeepSWE, the cyber surprise, open weights, pricing, and limits.

Gemini 3.7 Flash — Three-Week Iteration, Real Coding Gains, Now in Floatboat

Gemini 3.7 Flash ships three weeks after 3.6 Flash with a 50% introductory price cut and real coding gains. Benchmarks, the January price trap, and built-in access.

Grok 4.6 — Agentic Frontier Model, Benchmarks, Reality Check

Grok 4.6 explained: 500K context, $2/$6 pricing, the 200K token price trap, benchmark claims vs independent scores, and where it fits next to DeepSeek V4 Pro.

Claude Code vs Cowork vs Tag — Which Agent Surface Fits Your Work

Claude Code, Cowork, and Tag are three surfaces for the same agent: local coding, local office work, and shared async coworker. Pick by horizon and audience.

Best Claude Code Alternatives — Coding Agents Ranked by Job Shape

Ranked by coding job fit: Cursor, Cline, Aider, Devin, Codex CLI, Windsurf, and Copilot — and when Claude Code is still the right terminal agent.

What Is Claude Code — Anthropic's Terminal Coding Agent

Claude Code explained: Anthropic's terminal coding agent, its plan mode, sub-agents, and MCP — and how it differs from Cowork, Tag, and Chat.

What Is Claude Cowork — Anthropic's Agent for Knowledge Work

Claude Cowork explained: how it differs from Chat and Claude Code, what it does with local files and connected apps, and when knowledge workers should use it.

Best Claude Cowork Alternatives — Office Agents for Non-Developers

Ranked by job fit for non-developer knowledge work: calendar-driven agents, M365 Copilot Cowork, cloud agents like Manus and Perplexity Computer, plus Eigent and Genspark.

Best Claude Tag Alternatives — AI Teammates Ranked for Slack and Group Chat

Ranked list by team job fit: FloatIM for agent-native group chat, plus Viktor, Stilla, Runbear, Junior, and Operant — and when Claude Tag remains the best Slack-native choice.

Cordis — The Plugin Kernel Behind DeepSeek Harness

Cordis is the plugin framework powering DeepSeek Harness: reversible effects, reactive dependencies, and why 'everything is a plugin' matters for agents.

DeepSeek V4 Pro 0813 — The Quiet GA and What It Actually Changes

DeepSeek V4 Pro 0813 went GA with no announcement. Specs, claimed agent benchmarks, the pricing window before the hike, and how it stacks against Grok 4.6.

Grok Bot — xAI's Always-On AI Teammates, Explained

Grok Bot gives each AI teammate its own cloud computer that signs into your apps and works 24/7. Architecture, the security debate, pricing, and where it fits.

What Is MiniMax H3 — The Open Omni-Modal Video Model

MiniMax H3 is an open-weight omni-modal video model that turns text, images, clips, and audio into 2K video with native stereo sound. Capabilities and limits.

What Is a DeepSeek Agent? Types, Tools, and How to Choose

A DeepSeek Agent uses a DeepSeek model as its primary reasoning engine. Learn the four archetypes — native CLI, generic harness, chat assistant, and desktop — and how to pick the right one.

DeepSeek Agent vs Claude Code: Cost, Performance, and When to Use Each

Compare DeepSeek Agent vs Claude Code on cost, benchmarks, and setup. V4 Pro is ~28x cheaper on output — Claude Code still wins repo-scale refactors.

DeepSeek Agent Function Calling: A Hands-On Guide

Learn DeepSeek function calling for agents: tool schemas, strict mode, parallel tool calls, thinking mode, MCP, and production repair patterns — with runnable Python examples.

How to Build a DeepSeek Agent: From API Key to Your First Agent

Step-by-step guide to building a DeepSeek Agent with V4 Pro and V4 Flash. Covers API setup, model selection, tool calling, the agent loop, thinking mode, and production error handling — with Python and Node.js examples.

What Is Claude Tag — Anthropic's AI Teammate for Slack

Claude Tag explained: how @Claude works as a shared Slack teammate, how it differs from Cowork, Code, and Slack AI, and when teams should use channel-native agents.

How to Use Obsidian with an AI Agent — Local Vault + Cross-App Workflows

Learn how to use Obsidian with an AI agent: keep your vault as local Markdown, let a desktop agent read it across apps, tidy messy folders on-device, and run project work without leaving notes locked inside one room.

What Is an Obsidian Vault — Local Markdown Knowledge Base Explained

An Obsidian vault is a folder of Markdown notes on your device. Learn how vaults work, what they contain, how they differ from Notion, and why local files matter for AI.

What If Your Calendar Could Run Itself? Calendar-Driven AI

See how Calendar-Driven AI turns scheduled events into executable workflows, prep packages, drafts, and follow-up-ready work before the event starts.

Best Manus AI Alternatives in 2026

Looking for Manus AI alternatives? Here are the tools worth considering in 2026 — compared by use case, not just feature lists.

Nova