Skip to content
Zod Agent

The Zod Agent workspace.

A client-side, agentic AI chat workspace that runs entirely in your browser. Zero backend, no account, and it works with any OpenAI-compatible API.

Make it yours. Your sessions, memory, tools, provider keys, and preferences all live in your browser — nothing runs on our servers.

A chat workspace that feels like home

Start on the hub with a large prompt box and quick actions. Stream tokens as they arrive, switch sessions from a searchable drawer, and act on any reply — copy, regenerate, delete, or rate it.

Get started in minutes

Hub & sessions

A home screen, greeting, current time, and a big prompt box. Create, search, and switch sessions — search matches titles and message content.

Streaming chat

Token-by-token streaming with an auto-resizing input, a context-length meter, and message counts for every session.

Message actions

Copy, regenerate, delete, upvote, or downvote any assistant reply. User messages can be edited, copied, or removed.

Superpowers in the chat bar

Type / for slash commands, @ to mention agents, sessions, and tools, attach text files, or hit the schedule button to run the same prompt later.

Explore the docs

A prompt box that does more than type — command templates, mentions, attachments, and scheduling, all with keyboard navigation.

/ slash commands @ mentions file attachments schedule context meter

A real tool loop, running in your browser

When you send a message, Zod calls your provider with the full set of tool schemas, executes requested tool calls locally, and feeds results back — up to 6 rounds. Nine built-in tools, plus your own JavaScript tools and MCP servers over HTTP.

memory

Long-term markdown memory across sessions.

todo

A persistent to-do list, checked and tracked.

calculator

Safe arithmetic with a hand-written parser.

clipboard

Read and write the system clipboard.

timer

In-browser reminders with notifications.

code_tools

JSON & YAML formatting and conversion.

confirm

Ask before destructive actions with a yes/no modal.

random

UUIDs, dice, numbers, and WebCrypto hashes.

time

Current time, timezone conversions, IANA zones.

+ custom JS tools & MCP over HTTP

Structured answers, rendered as cards

The model emits lightweight [section] markers and Zod turns them into rich, interactive cards — no plugins required.

See the markers

Thinking & findings

Collapsible thinking steps, colored status badges, and findings lists with success, info, and error dots.

Code & files

Syntax-labeled code blocks, file chips, rendered HTML, and inline bar-chart visualizations.

Plans & confirmations

Clickable suggestions, checklists, and an Approve / Reject plan card with a tool checklist before anything runs.

Configure everything as YAML

Providers, memory, custom tools, and MCP servers are all editable as live, syntax-aware YAML in the settings drawer — with an "ask AI to edit this YAML" button on every section.

Learn about YAML config
Settings drawer — 9 tabs

General

Model, system prompt, request tuning, logging.

Providers

Default & fallback, retries, rate limits, timeout.

Memory

Persistent long-term memory as YAML blocks.

Tools

Custom JavaScript tools, editable and live.

MCP

Connect MCP servers over HTTP in a few lines.

Agents

List of agent harnesses, set active, edit.

Webhooks

Incoming triggers across all harnesses.

Schedule

Cron tasks — once, interval, daily, weekly.

Pairing

Device mesh sync with QR join and remote tools.

Orchestrate agent workflows

Compose multi-step workflows in a line-based editor — # title, ## @step:name, @tool:name, @skill:name, → @goto:name. Then run them, with per-step tool grants and live progress.

Learn about the orchestrator

A small workflow, ready to run:

# Research then summarize
## @step:gather
@tool:memory
@tool:time
→ @goto:summarize

## @step:summarize
@tool:todo
@skill:concise

Schedule tasks, even in the background

Run a prompt once, on an interval, daily, or weekly. Results land in a session you choose — and a service worker keeps executing due tasks even when the tab is closed.

Learn about the scheduler

Four modes

Once, repeat, daily, and weekly tasks — with run-now, enable/disable, and delete controls.

Delivery

Results land in a new session or one you pick — scheduled runs never disturb your chat.

Background agents

A service worker runs due tasks when the tab is closed, up to 3 at once, with notifications.

Make it yours.

No account. No server. Add an API key and your agent is ready — in your browser, as a Chrome extension, or as a desktop app.