You cannot switch update channels in Settings — channel follows the installed build.
You just opened kill9 and you're staring at panels, agents, and things called "initiatives." Here's what everything is and how it connects.
Overview
kill9 organizes your work into a hierarchy of entities, connects AI specialists (agents) to do the work, and uses workflows to make sure nothing skips a quality step. Agents remember what they learn, the chronicle records what happened, and extensions let you add new capabilities.
You don't need to memorize all of these at once. Start with initiative, story, agent, and session — the rest becomes useful as you go deeper.
What you can do
With these concepts, you can:
- Start a piece of work (initiative) and have agents handle planning, coding, testing, and review
- Break large features into manageable chunks (epics and stories)
- Chat with specialist agents who know exactly what role they play
- Keep agents consistent across sessions by giving them persistent memory
- Track every decision and action in a project timeline (chronicle)
- Add new tools and behaviors through extensions
How it works
Workspace
Your workspace is the full kill9 window. It has a left rail with seven surfaces — Inbox, Work, Plan, Docs, Mockups, Review, and Tools — and a main content area where you do your work. You switch surfaces with Mod+1 through Mod+7.
When you first set up kill9, you connect it to a project directory. That becomes your workspace. kill9 learns your codebase structure, build commands, and dependencies from it.
Initiative
An initiative is a major piece of work — a new feature, a bug fix, or a chore. It's the top-level container for everything related to that work.
You create an initiative when you have something meaningful to do. It has a type (feature, fix, or chore) and follows a workflow that moves it from idea to completion. Each initiative gets its own git branch (a worktree), its own documents, and its own chat sessions.
Example: "Add two-factor authentication" is an initiative.
Epic
An epic is a logical grouping within an initiative. Not every initiative needs epics — small fixes often go straight to stories. But for larger features, epics help you organize related stories.
Example: For the two-factor authentication initiative, you might have epics called "TOTP setup," "Recovery codes," and "Enforcement policies."
Story
A story is an atomic unit of work — the smallest thing that can be independently completed and verified. A story has a description, acceptance criteria, and a state. It belongs to an epic (or directly to an initiative if there are no epics).
Example: "Add QR code generation for TOTP enrollment" is a story within the "TOTP setup" epic.
Agent
Agents are AI specialists with defined roles, specific tools, and clear boundaries. kdev writes code. ktest verifies it. karch makes technical decisions. k9 orchestrates the whole pipeline.
You don't manage agents like a team lead. You describe what you want, and kill9 dispatches the right agent for the job. You can also talk to any agent directly in Work (Mod+2).
Session
A session is a conversation with an agent. Each initiative and story has its own session with full context and history. Sessions persist across restarts — close kill9, reopen it, and you pick up exactly where you left off.
Sessions carry the agent's working context: files it has read, commands it has run, and decisions it has made.
Workflow
A workflow is the path an initiative takes from start to finish. Each state has a purpose and designated agents. kill9 includes three workflow types: feature, fix, and chore.
Workflows prevent the common failure of jumping straight to code without clarifying requirements. The feature workflow, for example, requires an approved PRD before planning can start.
Gate
A gate is a quality checkpoint between workflow states. Gates block progress until evidence exists. The design gate requires an approved PRD and ADR. The planning gate requires stories with acceptance criteria.
Gates aren't bureaucracy — they prevent you from building the wrong thing or shipping broken code.
Document
Documents are reviewable artifacts attached to initiatives — PRDs (Product Requirements Documents), ADRs (Architecture Decision Records), design docs, and notes. They live in Docs (Mod+4) as a three-pane workspace with a navigator, editor, and agent chat.
Documents are part of the workflow. The design gate requires an approved PRD. Decisions get written down and reviewed before implementation starts.
Mockup
A mockup is a visual design created for review before coding starts. Mockups are HTML-based and render in the Mockups surface (Mod+5). You can generate multiple variants, compare them side by side, and select the winner.
Use mockups when the visual design matters — UI features, layout changes, or anything where seeing it beats reading about it.
Memory
Memory is persistent knowledge that agents retain across sessions. It includes your preferences, project conventions, architectural decisions, and any facts agents learn that should stick.
When an agent learns that your project uses Bun instead of Jest, that becomes a memory. Future agents see it automatically. You can also save memories manually.
See Memories for the full details.
Chronicle
The chronicle is your project's narrative timeline. It records what happened, when, and why — agent activity, state transitions, decisions, and outcomes. It synthesizes session activity into a readable story.
Use chronicle digests to catch up after being away, and chronicle search to find when a specific decision was made.
See Chronicles for the full details.
Skill
A skill is a composable behavior that augments what agents can do. Skills add specialized knowledge or workflow patterns — like a frontend expert skill that injects React best practices, or a security audit skill that adds OWASP checklists.
Skills are lighter than full agents. They layer on top of existing agents to specialize them further.
MCP
MCP (Model Context Protocol) connects external tool servers to your agents. When you connect an MCP server, your agents gain access to its tools — databases, APIs, browser automation, or anything the server provides.
MCPs extend agent reach without changing agent code.
Hook
A hook is a lifecycle automation trigger. When something happens in kill9 — a session starts, a tool runs, an initiative changes state — a hook can fire an automated response.
Built-in hooks handle things like memory extraction and chronicle recording. You can create custom hooks for your own automation.
See Hooks for the full details.
Worktree
A worktree is an isolated git branch workspace for an initiative. Each initiative gets its own worktree, so multiple initiatives can progress in parallel without stepping on each other.
You can switch between worktrees without losing work. This matters when you're juggling a feature, a hotfix, and a refactor at the same time.
Channel
Channels are release tracks: alpha, rc (release candidate), and stable. Alpha gets features first but may be rough. Stable gets them last but is the most reliable. You pick your channel in settings.
If you want early access to new capabilities, use alpha. If you need reliability above all else, use stable.
Tips
- You don't need to understand every concept before starting. Create an initiative, start chatting, and learn the rest as you go.
- The entity hierarchy is initiative → epic → story. Think of it as project → phase → task.
- If you're unsure which agent to use, just talk to k9. It delegates to the right specialist.
- Memory quality directly affects agent quality. Spend a few minutes reviewing and cleaning up memories periodically.
- Worktrees are automatic — you don't manage them. Just know that each initiative's code changes are isolated.
Related
- Agents — the full list of agents and when to use each
- Workflows — how initiatives move through states
- Documents — writing and reviewing PRDs and ADRs
- Memories — persistent knowledge across sessions
- Chronicles — your project's narrative timeline
- Hooks — lifecycle automation
- Extensions — adding capabilities through skills, MCPs, and plugins