zot

What it does

Writes code

YesThe built-in toolbox is read, write, edit, glob and bash, with write creating or overwriting files directly.Five built-in tools. Zero ceremony. The minimum viable toolbox for an agent that actually ships code.zot.sh/docs/tools ↗

Debugs and fixes

PartlyNo dedicated debugging feature; the vendor names failure investigation only as an example of a task you can hand a background sub-agent, which then edits the same working directory.Spawn one for a side task (“draft the migration”, “investigate this stack trace”, “write the test harness for module X”), keep going in the main…github.com/patriceckhart/zot ↗

Writes tests

PartlyTest authoring appears only as an example task for a swarm sub-agent; there is no test-generation feature in the tool set or the docs.Spawn one for a side task (“draft the migration”, “investigate this stack trace”, “write the test harness for module X”), keep going in the main…github.com/patriceckhart/zot ↗

Reviews code

PartlyReview is not in the binary: it comes from a separately installed extension by zot's own author, listed in the vendor's extension directory.Structured repo-wide code review for zot.zot.sh/extensions ↗

Whole-codebase work

PartlyRepo-wide reach is recursive glob plus bash; the documented tool set is five tools and none of them indexes a codebase.Searches recursively, honors root and nested .gitignore rules, excludes hidden paths by default, and respects jail mode.zot.sh/docs/tools ↗

Ships it

PartlyShipping is whatever the shell can do: bash runs commands in the session working directory. No deploy, pull-request or git-forge integration is documented.Run a command in the session cwd with merged stdout/stderr and a timeout.zot.sh/docs/tools ↗

How much it does unattended

Runs autonomously

YesTools run without confirmation by default; the confirmation flag exists only in the TUI and is deliberately ignored in print, JSON and RPC modes.In -p, --json, and rpc modes it is ignored with a stderr warning and tools still run freely, so scripts and automation keep working.zot.sh/docs/tool-gate ↗

Multi-agent

Yes/swarm spawns background subagents, each its own subprocess and model loop; with auto-swarm on, the main agent can fork them itself via a swarm_spawn tool.Each agent is a separate zot subprocess with its own model loop, its own persistent session file, and its own chat in the dashboardzot.sh/docs/swarm ↗

Agent permissions

YesFree-running by default, with three brakes: --no-yolo confirms every call in the TUI, /jail confines tools to the working directory, and a zotfile manifest caps file and bash access.Run with --no-yolo to confirm every tool call before it runs.zot.sh/docs/tool-gate ↗

Which models it runs on

Claude

YesClaude runs through an Anthropic API key, a Claude Pro/Max subscription login, or Bedrock and Vertex model IDs.Anthropic Claude Pro/Max (anthropic), OpenAI Codex / ChatGPT Plus/Pro (openai-codex), Kimi Code (kimi), and GitHub Copilot (github-copilot).zot.sh/docs/providers ↗

GPT

YesGPT and Codex models are baked into the catalog and reachable by OpenAI API key, a ChatGPT/Codex subscription login, or Azure OpenAI Responses.Built-in entries cover Claude, GPT/Codex, Gemini/Gemma, Kimi/Moonshot, DeepSeek, Groq-hosted models, OpenRouter-routed models, Bedrock and Vertex…zot.sh/docs/models ↗

Open models

YesOpen weights run locally through Ollama, an OpenAI-compatible endpoint, or a managed llama.cpp router with GGUF download, load and unload from /llama.Managed GGUF models: connect the distinct llama.cpp provider to a recent llama.cpp router, then use /llama for download, load, unload, and removal.zot.sh/docs/providers ↗

Model choice

YesYou pick: one merged catalog across every configured provider, switchable mid-session with /model or Ctrl+1..9, and extendable with your own models.json entries.--list-models and /model show the merged catalog across every provider.zot.sh/docs/models ↗

Where you use it

In your editor

NoA terminal agent. The documented run modes are the TUI, print, JSON and an RPC child process; no editor or IDE plugin is offered.zot is a lightweight terminal coding agent, shipped as one static Go binary. No runtime. No Docker. No plugin system with a dedicated package manager.zot.sh/docs ↗

On the command line

YesThe CLI is the product: an interactive TUI plus a one-shot print mode that reads piped stdin and writes final text to stdout.print — zot -p, one-shot, final assistant text to stdout. Great for shell pipelines.zot.sh/docs ↗

In your pipeline

PartlyJSON mode is pitched at CI and emits NDJSON events, but no pull-request, GitHub Action or forge integration is documented; you wire the binary in yourself.json — zot --json, NDJSON events to stdout. Perfect for scripts and CI.zot.sh/docs ↗

In a browser

NoNo hosted web app. The only remote surface documented is a Telegram DM bridge over a protocol-agnostic bot core.zot is a lightweight terminal coding agent, shipped as one static Go binary. No runtime. No Docker. No plugin system with a dedicated package manager.zot.sh/docs ↗

Whose machine it runs on

Self-hosted

YesSelf-hosting is the only mode: a static binary on your machine, using your credentials, in your working directory. There is no vendor-run backend.The user runs it locally with their own model credentials, in their own workspace, using zot as the runtime.zot.sh/docs/zotfiles ↗

Open source

YesOpen source, and the footer of every page states it is MIT licensed and built in Go; source and docs are published on GitHub.This website provides information about zot, an open-source coding agent harness.zot.sh/disclaimer ↗

What it costs to run

How it meters

YesNothing is metered by zot: the binary is free and every token is billed to the API key or subscription you attach with /login.Use /login to store an API key or a subscription credential.zot.sh/docs/quickstart ↗

Free tier

YesOne install command on macOS, Linux or Windows, or go install from source. No account with the vendor exists to sign up for.One command on macOS, Linux, or Windows.zot.sh ↗

API access

YesThree ways in: a Go in-process SDK, zot rpc speaking NDJSON over stdin/stdout, and a packaged Node.js TypeScript client plus a Swift xcframework.Spawn zot rpc as a subprocess and exchange NDJSON over its stdin/stdout. Reference clients for Python, Node, shell, and Go live under examples/rpc/.zot.sh/docs/embedding ↗

MCP server

PartlyMCP is deliberately not built in; an extension proxies configured MCP servers into zot. Nothing documents zot serving MCP to another agent; that path is the RPC protocol.Build CLI tools with skills, or add an extension that speaks MCP if you actually need it.zot.sh ↗

Bring your own key

YesBring your own everything: provider API keys, existing Claude/ChatGPT/Kimi/Copilot subscriptions, keys fetched from a password manager, or your own AWS, Azure, Vertex or local endpoint.zot supports Claude Pro/Max, ChatGPT/Codex, Kimi Code, and GitHub Copilot subscriptions, plus direct API keys for every provider in the catalog.zot.sh/docs/quickstart ↗

Buying it for a team

Admin controls

PartlyConstraints are per-machine, not centrally administered: a zotfile manifest caps file and bash access, and jail_by_default in config.json starts every run confined to the working directory.zotfile permissions are an enforced ceiling for the built-in file and bash tools. Empty or omitted filesystem scopes deny access.zot.sh/docs/zotfiles ↗

Audit log

PartlyEvery session keeps a full local JSONL transcript, inspectable as a timeline and exportable as JSON, but it is a user-owned transcript rather than an administrative audit trail.Every turn is appended to a JSONL transcript under $ZOT_HOME.zot.sh/docs/sessions ↗

Single sign-on

NoAuthentication is provider credentials only: subscription OAuth or API keys. No SAML or OIDC for zot itself.Use /login for subscription auth and simple API-key providers.zot.sh/docs/providers ↗

What happens to your code

Data residency

PartlyYou choose the jurisdiction with the provider: Bedrock runs against the AWS_REGION you set and some models need us. or eu. regional inference-profile IDs.Some Bedrock models require regional inference-profile IDs for on-demand throughput, such as us. or eu. prefixed model IDs.zot.sh/docs/providers ↗

Getting out

YesTranscripts are plain JSONL files you own, /session export writes a portable .zotsession and the timeline exports to JSON; swarm subagent state does not round-trip. No credits exist to expire./session export writes the transcript to a portable .zotsession file. /session import pulls one back in as a first-class resumable session.zot.sh/docs/sessions ↗

Not established: Opt out of training, Certifications