Kit by Speakeasy Review — A Concise, One-Tool Coding Agent

What Is Kit by Speakeasy?

This is a coding agent runtime, released open source by Speakeasy in September 2026. It rethinks how agent tools work by giving the model exactly one tool — a single compose interface — instead of a sprawling toolbox of model-specific functions.

The project was announced as a Show HN on Hacker News in early September 2026, where its author framed it as a concise alternative to heavyweight agent CLIs. It is not the same project as ACP Kit, a separate Agent Client Protocol implementation, nor mark3labs/kit, a knowledge-inference tool — Kit by Speakeasy is its own runtime.

A single binary ships the whole experience, which makes installation, upgrades, and CI usage straightforward.

One Tool: Compose and the Runlet Language

The design assumption is that most agent tools are just composed calls: fetch a file, run a command, read the output, and iterate. The runtime therefore exposes a single compose tool, written in a custom language called Runlet that is optimized for tool-call composition.

By constraining the model to one well-defined interface, the runtime keeps agent intent explicit and auditable. The model composes smaller operations rather than invoking ambiguous mega-tools, which the project argues produces more predictable behavior.

Speakeasy maintains Runlet as its own project (github.com/danielkov/runlet), and Kit embeds it as the execution language for that one tool.

• Exactly one tool: compose
• Runlet language built for tool-call composition
• Agent intent stays explicit and auditable
• Maintained as github.com/danielkov/runlet

Terminal, ACP, and A2A in One Binary

The runtime runs in three modes from the same binary. The terminal client (TUI) is the primary surface for interactive work, while an Agent Client Protocol (ACP) server lets Kit plug into compatible editors, and an A2A endpoint exposes the agent for peer-to-peer agent communication.

ACP is important because it decouples the agent from any single editor: Kit can run inside an ACP-compatible editor and can also orchestrate other agents such as Claude, Codex, or Cursor as subagents.

The bundled subagent orchestrator means a single kit process can coordinate nested work, and because everything is in one static binary, there are no separate daemons or dependency trees to manage.

• TUI for interactive terminal use
• ACP server for editor integration
• A2A endpoint for other agents
• Subagent orchestration built in

Installation and Quick Start

The README quickstart is deliberately short. Install by piping the official installer: curl -fsSL https://raw.githubusercontent.com/speakeasy-api/kit/main/install.sh | sh. Then run kit init, kit auth login openai, and kit tui to authenticate with an OpenAI-backed provider and start.

A container image is published at ghcr.io/speakeasy-api/kit, so the same workflow runs in isolated or CI environments.

The project tracks releases with semantic versioning tags, and its README shows release badges, which helps teams pin a known-good version.

• curl installer from the official repo
• kit init; kit auth login openai; kit tui
• Container at ghcr.io/speakeasy-api/kit
• Semantic versioning release tags

Extending Kit: Skills, Plugins, and MCP Servers

The runtime reloads configuration dynamically, so skills, plugins, and MCP servers can change without a full restart. That fits an agent runtime that is meant to be tuned continuously.

The dynamic reload model lowers the friction of testing new skills or swapping tool servers during a session, which matters for long-running agent work.

Because the runtime speaks ACP and A2A, extension also happens at the agent level: Kit can appear as a peer or subagent to other ACP-compatible tools rather than only executing in isolation.

• Dynamic reload of skills
• Plugins and MCP servers reload without restart
• ACP- and A2A-compatible at the agent level

Pricing

Kit by Speakeasy is open source and free to use. There is no hosted tier or license fee described in the README — you pay only for the underlying model API, which the kit uses through your authenticated provider login.

For a team comparing agent runtimes, that makes it effectively a bring-your-own-model client: zero software cost against whichever frontier model you log in with.

The absence of a managed platform means setup, security, and updates are your responsibility, which is the standard trade-off for self-hosted agent tooling.

• Open source and free
• Pay only for model API usage
• Bring-your-own-model provider login
• Self-hosted, so ops are yours

Pros and Cons

The runtime wins on simplicity and portability: one binary, one tool, three interfaces, and a clean install path. The one-tool design is genuinely different from mainstream agent CLIs, and dynamic reload keeps iteration fast.

The caveats are its age and reach. The project is days old, has a small community, and requires learning a custom composition language. There are no verified user ratings on this directory yet, so treat this review as a technical overview rather than long-term field evidence.

Best For

Recommended use cases and scenarios where Kit by Speakeasy shines.

Frequently Asked Questions

Common questions about Kit by Speakeasy, answered.