Zot

What is Zot?

Zot is a single-binary coding-agent harness written in Go by independent developer Patrice Eckhart, who describes it, self-deprecatingly, as "yet another coding agent harness." It runs as a terminal TUI, a scriptable print/JSON-streaming mode, or a headless daemon, connecting to 30+ model providers, Anthropic, OpenAI/Codex, Gemini, Kimi, DeepSeek, GitHub Copilot, Bedrock, OpenRouter and local Ollama models among them, through one static binary with no Node runtime, Docker or package manager required.

What does Zot do?

Point Zot at a repo and it works through four built-in tools — read, write, edit, bash — while you drive it from a terminal TUI, a scripted print mode, or JSON streaming for tool integrations. Its most distinctive feature is session control: /session fork copies every message up to any past turn into a brand-new session, so you can branch an approach without losing the original thread, and /session tree renders every session in the current directory as a parent/child tree so you can see the branches at a glance — a CLI-first take on the same "conversation as a tree, not a transcript" idea Juggler ships as a GUI. A Telegram bridge lets a session run in-TUI with DMs mirrored bidirectionally, or as a standalone headless bot (zot telegram-bot) you message from a phone. "Swarm" spins up background subagents in parallel on the same repo, and an extension system (subprocess + JSON-RPC, any language) adds custom tools and slash commands without touching Go code. It also ships as an importable Go package (packages/agent/sdk), so other developers can embed Zot's agent loop in-process to build their own harness rather than using it only as an end-user CLI. The project ships extremely fast — 303 releases and counting since an April 2026 start — and its own docs describe it as "lightweight and in beta forever," not a claim of production stability.

Key features

• Fork-from-any-message + /tree view: /session fork branches a new session from any past turn; /session tree renders every session in the working directory as a parent/child tree.
• 30+ model providers, one static binary: Anthropic, OpenAI/Codex, Gemini, Kimi, DeepSeek, GitHub Copilot, Bedrock, Azure OpenAI, OpenRouter and local Ollama models, with no Node runtime or Docker needed to run it.
• Telegram bridge: Mirrors an in-TUI session to Telegram DMs bidirectionally, or runs as a standalone headless bot (zot telegram-bot) for driving the agent remotely from a phone.
• Swarm (background subagents): Runs multiple subagents in parallel on the same repository instead of one linear session at a time.
• Importable as a Go SDK: The agent loop is published as packages/agent/sdk, so other developers can drive Zot in-process from their own Go program rather than only via the CLI, positioned for building your own harness, not just using this one.

What are Zot's use cases?

• Branching exploration mid-task: An agent goes down the wrong path partway through a coding session; the developer forks from the last good message, tries a different instruction in the new branch, and keeps both threads via /session tree for comparison.
• Remote/mobile supervision: A developer kicks off a long-running task from a desktop terminal, then messages the Telegram bridge from a phone to check progress or approve a step without opening a laptop.
• Embedding an agent loop in a custom tool: A developer building their own internal coding-agent product imports packages/agent/sdk directly into a Go program instead of writing a provider-routing and tool-execution loop from scratch.

What does Zot integrate with?

• Anthropic, OpenAI/Codex, Gemini, Kimi, DeepSeek, GitHub Copilot, Bedrock, Azure OpenAI, OpenRouter, Ollama (local models)
• Telegram (in-TUI bridge or standalone bot)
• JSON-RPC extensions (any language)
• Go SDK (`packages/agent/sdk`, in-process embedding)

Why use Zot?

• Free and MIT-licensed (a permissive license, unlike some rival open-source harnesses' copyleft terms), a single static Go binary, no signup, no Docker, no Node runtime.
• A real, working fork-from-any-message + tree-view session model, not just a linear transcript, genuine differentiation the HN launch thread and independent reviews both confirmed works as described.
• Written entirely in Go rather than TypeScript/Node, which several Hacker News commenters specifically said was the reason they wanted to try it over JS-based rivals.
• Actively and rapidly shipped: 292 GitHub releases and counting since its ~April 2026 start, with 10+ external contributors beyond the original author, not a one-person weekend project left to stall.

Pros & cons

Zot pricing

• Open sourceFree

See current pricing on zot.sh ↗Compare Zot alternatives →

Zot review

Zot earns its place with a real, working idea: fork any past message in a coding session into a new branch and see the whole tree at a glance, all from one free, MIT-licensed Go binary with no Docker or Node runtime. It backs that up with genuine breadth — 30+ model providers, local Ollama support, a Telegram bridge, background swarm subagents, and an importable Go SDK for developers who want to build their own harness on top of it — and it ships fast, with 303 releases and 10+ contributors since an April 2026 start. The honest catches are the ones the maintainer discloses directly: the OAuth-reuse trick that lets it bill through a Claude/ChatGPT subscription is, by the README's own admission, a likely terms-of-service violation that could be revoked at any time (the API-key flow sidesteps this), the docs call the project "beta forever" rather than production-stable, and the TUI is admitted to be almost entirely AI-generated. Pick Zot if you want a free, fast, Go-native harness with real branching-session control and are comfortable with fast-moving independent software; pick Juggler instead if you want the same branching-tree idea with a GUI, or a first-party vendor tool if you need managed support and want to avoid any OAuth-reuse risk entirely.

A genuinely fast-moving, CLI-first take on branching sessions, the fork + tree view works as advertised, but read the maintainer's own OAuth and "beta forever" disclosures before adopting it for anything you cannot afford to have break.

Frequently asked questions