CORE FEATURES

A complete Agent platform — everything in one place

Web UI, CLI, long-term memory, Skill library, scheduled tasks, IM integration, browser automation. The full day-to-day AI workflow, delivered in one product — and every piece runs on the model you choose.

Install now Read full docs
WEB UI

Start locally. Use it from any browser.

Launch the local server and open it in your browser — sessions on the left, conversation in the middle, artifacts on the right. Run multiple tasks in parallel, revisit history at any time, no commands to memorize.

Run tasks in parallel

Each task gets its own session. Work in parallel without interruption — review whichever finishes first.

No commands required

Switch sessions, pick skills, change models — all through the UI. Nothing to memorize.

Everything archived

Conversations, artifacts, and tool calls are persisted automatically — resume any time, nothing is lost.

CLI

A CLI — the same capability, for terminal users

Type openclacky in your terminal and you're in — no server, no login. Shares configuration and memory with the Web UI. The open-source alternative to Claude Code.

Install and go

No daemon, no cold start. One gem install and you can start working in any project directory immediately.

Three-tier permissions

From step-by-step confirmation to full autonomy — three switchable levels. Diff preview before writes, guardrails on destructive operations.

Project rules auto-loaded

Reads the directory's .clackyrules on startup. The Agent immediately understands your stack, conventions, and common commands.

Unique capability Rarely found in similar products

Skill auto-evolution — the library grows with use

After each task, OpenClacky evaluates whether anything in this round is worth preserving. If so, it either writes a new Skill or improves the one that was just used — all without interrupting the current conversation.

Subagent Label
1
Step 1 Label
Step 1 Html
2
Step 2 Label
Step 2 Html
3
Step 3 Label
Step 3 Html
Step 4 Label
Step 4 Html
~/.clacky/skills/ auto-update
.
├─ deploy/
│  └─ SKILL.md
├─ commit/
│  └─ SKILL.md
├─ browser-setup/
│  └─ SKILL.md
├─ meeting-notes-to-tasks/  just created
└─ SKILL.md  (written by skill-creator)
Case one
Repeat work, automatically distilled into a Skill
A task completed without any existing Skill, guided step by step by you. If the workflow is clear and reusable, the system captures it as a new Skill — available via slash command next time.
Case two
Existing Skills sharpen with every use
When an existing Skill meets a case it hadn't seen before, the system folds that handling back into the Skill — making the next invocation more stable and better aligned.
Only user-created Skills (under ~/.clacky/skills/) are modified; built-in Skills are never touched. Runs silently in the background and can be disabled in settings at any time.
How the mechanism works
SKILL LIBRARY

Skill · installable, reusable, shareable capability units

A Skill is a capability unit in OpenClacky — a curated workflow paired with usage instructions. A default set ships with the product; community skills install with a single command, and you can author your own and publish them to the marketplace.

Installed skills 12
/skill-add https://...
C
/commit
Analyze diff and generate a semantic commit
D
/deploy
One-click deploy to Railway
B
/browser-setup
Configure Chrome / Edge automation
P
/pptx
Generate a PowerPoint from scratch
P
/personal-website
Generate and publish a personal homepage
C
/cron
Create and manage scheduled tasks
S
/skill-creator
Turn a workflow into a Skill
S
/skill-add
Install a community skill from a URL
S
/stripe-webhook-debug
Break meeting notes into assigned tasks
MINE
One command to install any community Skill

Ready to use

commit, deploy, browser-setup, pptx, personal-website and other first-party Skills — available via slash commands immediately after install.

Simple to author

Use skill-creator to wrap any workflow into a Skill: describe the purpose, outline the steps, specify when to trigger. Typically done in minutes.

Community & marketplace

Install from any URL with /skill-add. Publish polished Skills to the marketplace — share for free or monetize via revenue share.

LONG-TERM MEMORY

Memory · key context captured, recalled on demand

Key decisions, architecture choices, and preferences from your conversations are extracted and persisted automatically. When the topic comes up again, the relevant memories are recalled without any manual action.

~/.clacky/memories/
.
├─ acme-delivery.md
├─ customer-commitments.md
├─ team-preferences.md
├─ tech-stack-choices.md
├─ db-schema.md
└─ deployment-notes.md
In conversation · auto-recall
How did we plan the delivery milestones for this customer?
Matched acme-delivery.md
acme-delivery.md hit

Acme project · delivery milestones

Three-phase delivery: Phase 1 (5/15) requirements lock-in and prototype; Phase 2 (6/10) MVP handover and UAT; Phase 3 (6/30) production launch. Key dependency: customer-side SSO integration must be done by 5/20. Weekly sync every Friday; milestone risks are communicated by Mia to the customer.
Memory files are stored locally in ~/.clacky/memories/. The Agent determines when to recall — no manual trigger required.

Captured automatically

Key decisions, technical choices, and personal preferences are extracted and written to local Markdown files on their own.

Recalled by relevance

Memory is not dumped into every prompt — it is loaded only when judged relevant to the current topic. Saves tokens and improves accuracy.

Fully local

Memory files live in your own ~/.clacky/memories/ — human-readable, editable, and version-controllable.

AUTOMATION & INTEGRATIONS

Bring the Agent out of the chat window, into real workflows

Trigger on schedule, summon from IM, operate real pages in the browser — plug the Agent into the tools your team already uses, and let it deliver value in the background.

Scheduled tasks 4 active
+ New task
Task Schedule Next run Status
Daily AI news digest
0 9 * * * Tomorrow 09:00 Active
Monitor prod error logs
*/30 * * * * in 15 min Active
Friday weekly report draft
0 18 * * 5 Fri 18:00 Active
Heartbeat check
0 * * * * Paused
Describe what you need in natural language — OpenClacky generates the cron schedule. Start, pause, and inspect run history in the Web UI.
IM integration

Feishu / WeCom / WeChat — invoke it with @

Connect the Agent to the IM platforms your team already uses. No context switching — mention it or DM it. Each conversation thread maps to its own isolated Session.

# product-eng 12
Y
You · now
@OpenClacky Count this week's new users and post the summary here
C
OpenClacky · just now
1,248 new users this week (+18% WoW)
— Web 763
— iOS 312
— Android 173
Browser automation

Drive Chrome / Edge to complete real web tasks

The Agent can read pages, click buttons, fill forms, and download files — handling work that requires a real browser environment. One-line start via browser-setup.

github.com/clacky-ai/openclacky
clacky-ai / openclacky
Your personal Assistant & Technical Co-founder · CLI + Web UI + Skill Platform
☆ Star
browser.navigate "https://github.com/clacky-ai/openclacky"
browser.click ref=e8  → star button
ENGINEERING & INTERFACE

The engineering details that power users notice

From project rule files to BYOK, from idle-time compression to fine-grained permissions — built for keyboard-first power users, and for teams that need shared conventions.

Project rules

.clackyrules · project-level conventions, auto-loaded

Place a .clackyrules file in the project root declaring stack, conventions, and commands. Every time the Agent starts in that directory, it reads and follows them.

.clackyrules auto-loaded
# Project Rules

## Stack
- Ruby 3.3 + Rails 7.2
- PostgreSQL 16, Redis 7
- TailwindCSS, Stimulus, Turbo

## Conventions
- IMPORTANT: All new code in TypeScript — avoid any
- Controllers handle HTTP only; business logic lives in Service
- PRs must pass rubocop + rspec before merging

## Commands
- Dev: bin/dev
- Test: bundle exec rspec
- Lint: bundle exec rubocop -a
BYOK

Bring your own API key — choose any provider

OpenClacky, Anthropic, OpenRouter, DeepSeek, Kimi, Minimax, GLM, MiMo, and any OpenAI-compatible endpoint — switch with one click. We do not proxy, do not store keys, and do not take a cut.

Pick a model provider 8 built-in
OpenClacky abs-claude-sonnet-4-5
Anthropic claude-sonnet-4.6
OpenRouter Multi-provider aggregator
DeepSeek V4 deepseek-v4-pro
Kimi kimi-k2.6
Minimax MiniMax-M2.7
GLM glm-5.1
MiMo mimo-v2-pro
Custom endpoint openai-compatible

Cost under control

Idle-time compression combined with precise context injection — roughly 20% lower cost than Claude Code for equivalent workloads, measured.

Smart compression

Conversation history is compacted in the background between sessions. When you return, the tokens loaded into the model are already minimized.

Complete toolkit

File I/O, shell execution, web search, code search, browser, IM — 10+ built-in tools, ready out of the box.