Lively

The agent OS where your team and its AI work from the same context.

한국어 · Website · Managed (app.lvly.io) · Architecture

Lively has AI collect your team's documents, conversations, code, and decisions, keep only what is still true, and feed as much of it as needed to whichever AI each teammate uses (Claude Code, Codex, and others). On top of that context, agents do the work, and the apps your team builds read from the same context.

In an OS	In Lively
Memory	Context store: collection, duplicate and outdated-fact checks, injection. Session output flows back into the store
Processes & shell	Workspace: projects, tasks, and sessions that connect into one flow. Developers use the terminal; everyone else uses the web or desktop app
Apps	Apps: agents and tools your team builds run on the same context
Drivers	Integrations: Slack, Notion, GitHub, databases, and more

The core is open source under AGPL-3.0 and free to use with no limit on team size. If you'd rather not run a server yourself, use the managed service.

---

Why

Teams now use much the same models. The difference in results comes from whether the AI knows your team's work. Yet nobody is looking after that context.

• You find things, but can't tell what's right. A two-year-old spec and last week's changed decision come back with equal weight, and the AI reads the old one and is confidently wrong.
• What you organize goes stale. You build a wiki, and nobody keeps it up to date.
• Work done with AI disappears when the session ends. A conclusion one person reached with AI today is unknown to someone else's AI tomorrow.
• The best setups live on one person's machine. The one or two people who built skills, hooks, and MCP integrations hand real work to AI; everyone else uses it as a search box.

Lively hands that upkeep to AI instead of a person.

What's inside

Context store

• Collect: Reads from the tools you connect, read-only. Instead of piling up raw material, it keeps the facts and decisions as knowledge.
• Judge: Merges content that already exists and replaces decisions that have changed with the latest version. Every piece of knowledge carries its source.
• Organize: Links knowledge to your taxonomy and projects. Turn on embeddings for semantic search.
• Inject: Puts relevant knowledge into the AI when a session starts or a question is asked. Only what matters goes in full; the AI looks up the rest over MCP when it needs it.
• Write back: Conclusions the AI reaches in a session come back as projects and knowledge for the next person's AI to use. Duplicates are caught before saving, and people review afterwards.

Shared setup

Promote a skill, hook, subagent, or MCP integration that one person built into an organization asset, and it is installed automatically in every teammate's AI. You can roll it out company-wide, per team, or per person, and new hires start in the same environment on day one.

Workspace

• Projects and tasks: Work flows through projects and tasks, and each AI session takes one task. Progress and the knowledge produced stay in one place, with no separate PM tool.
• Web terminal: Open an isolated AI session in the browser. Work files remain after the session ends. Teammates who aren't at home in a terminal work from the same context.
• Your machine and the desktop app: Connect Lively to a local Claude Code (or another AI tool), or use the desktop app for macOS, Windows, and Linux.
• Internal database queries: Connect a data source such as a production read replica, read-only, and people can ask the AI for data without going through the engineering team. The SQL firewall and table allow/deny policy apply.
• Scheduled runs: Run AI jobs at set times.
• Liv: An AI assistant for first-time setup. It looks at your team's material and work, proposes which tools to connect and how to configure them, and applies the changes once you approve.

Apps

Install apps (agents and tools) your team builds into the workspace, and they run inside your organization's context and permissions. App installation, permissions, and the SDK are included; a marketplace for installing apps built by other organizations is in progress.

How it compares

	Good at	How Lively differs
Built-in memory and rules files in AI tools (CLAUDE.md, etc.)	Works right away with nothing to install; strong for per-repo rules	People have to write and maintain it, and it stays per person or per repo
AI subscription work agents	Start inside a subscription you already have, with no setup	Memory is per person, and session output doesn't accumulate for the team. Works only inside that vendor's AI
AI in document tools	Excellent editing, search, and summarization of documents	AI-generated documents pile up without review or dedupe. Code and databases sit outside the tool
Assemble it yourself (memory SDK + MCP server + hooks)	Free choice of components, no license cost	You integrate collection, judgment, permissions, and UI yourself, and keep operating it

Lively sits on top of the AI tools you already use, without replacing them. Change models or tools, and the context stays in Lively.

You may not need Lively if:

• You work alone. Lively's value grows when several people and several AIs share the same context.
• You use a single AI tool, and its memory and rules files are enough.
• You have a platform team to run a stack you assembled yourself, and not much knowledge to accumulate.

This is the beginning of the README. Read the rest at the original. Read the full README ↗