Slate AI

Slate (21) akira (@realmcore_) / X

🧠 What Is Slate?

Slate is an AI-powered workspace designed for long-context reasoning and agentic workflows. It blends coding, note-taking, project management, and multimodal input (text, speech, images) into a single interface.

⚙️ Key Features

• Ultra-long context support: Handles up to 1 million multimodal tokens, which is massive—think 750,000 words or 5,000 HD images in one query.
• Agentic workflows: Includes an AI agent called Polymath that can automate tasks like project outlining, web search, and more.
• Multimodal input: You can combine text, speech, and images in your queries. Output is currently limited to text and speech (TTS).
• Privacy-first design: Slate claims it never uses your data for training or sharing—ideal for pseudonymous or sensitive workflows.
• Developer-friendly: Supports advanced LLMs from OpenAI, Anthropic, Gemini, and even self-hosted DeepSeek models.

🧑‍💻 Coding Capabilities

While it’s not just a code generator like GitHub Copilot, Slate is built to support:

• Complex project outlines
• Code snippets and refactoring
• Integrated task and knowledge management
• Customizable agent instructions and response formatting

It’s still early days, but Slate is positioning itself as a powerful alternative to fragmented dev tools, especially for those who want privacy, context depth, and agentic control.

You can explore more on Slate’s official site. Want to compare it to your current stack or brainstorm how it could fit into your agent onboarding flow? I’d love to dive deeper.

2026.09.08

Status: active — but the product has pivoted from “long-context workspace” to an orchestration engine. Still Random Labs (YC S24; akira @realmcore_, “teaching agents how to do engineering”). Slate is now framed as a generalist software agent for parallel swarm orchestration — a scriptable runtime for execution graphs (fan-out → explore → evaluate), with subagents, budgets, model choices per workflow, and persistent work state across handoffs/restarts. Think of it as less “chat workspace with 1M tokens” and more “durable agent program runner” — e.g. /deepresearch as a running program rather than a task breakdown.