Init templates

kiwifs init creates a knowledge directory from a built-in template. Pick the template closest to how you will use the knowledge base. Everything it creates is plain markdown plus a .kiwi/ folder, so you can change any of it later.

Terminal

kiwifs init --template wiki --root ./knowledge
kiwifs serve --root ./knowledge

Flag	Default	Description
`--template`	`kb`	Template name (see below)
`--root`, `-r`	`./knowledge`	Directory to initialize

init never overwrites existing files, so it is safe to run on a folder that already has content.

Available templates

Template	Best for	What you get
`kb`	Curated knowledge bases	Article types, verification workflow, freshness checks
`wiki`	Team wikis	Onboarding, decision records, processes, reference docs
`memory`	Agent memory	Episodic and semantic pages, consolidation log
`tasks`	Task tracking	Task schema, status workflow with WIP limits, claim fields
`data`	Structured records	Collections, DQL dashboards, chart widgets
`cms`	Headless CMS	Blog, docs, pages, authors, editorial workflow, feeds
`runbook`	Ops and incident response	Severity and service frontmatter, execution staleness
`adr`	Architecture decision records	MADR format, status workflow, auto-numbering
`prompt`	Prompt registry	System and task prompts, eval rubrics
`research`	Literature and papers	Paper tracking, reading workflow, typed citation links
`log`	Append-only event logs	Daily files that reject overwrites
`blank`	Starting from scratch	Config, rules, and page templates only

The old template names knowledge and prompt-library are now memory and prompt. Using an old name returns an error that names the new one.

What every template includes

Every template, including blank, creates:

.kiwi/config.toml — server, search, versioning, and auth settings. Some templates ship their own; the rest get the default config.
.kiwi/rules.md — agent rules you can export with kiwifs rules export.
.kiwi/templates/ — page templates for decisions, meeting notes, onboarding, product specs, SOPs, and tasks.
.gitignore — ignores .kiwi/state/, which holds rebuildable indexes.
README.md

Every template except blank also includes two documents for agents:

SCHEMA.md — folder layout, frontmatter fields, and naming conventions.
playbook.md — step-by-step MCP tool sequences for common operations. init writes it to both the root and .kiwi/playbook.md. Agents receive it through kiwi_context.

See Agent playbook.

Export agent rules

Turn .kiwi/rules.md into the rules file your agent harness reads:

Terminal

kiwifs rules export --format cursor    # default
kiwifs rules export --format claude
kiwifs rules export --format agents
kiwifs rules export --format skill
kiwifs rules export --format openclaw

Template reference

kb — knowledge base

A curated knowledge base where articles are reviewed and kept fresh.

knowledge

SCHEMA.md

playbook.md

index.md

faq

getting-started

guides

reference

troubleshooting

.kiwi

config.toml

schemas/article.json

workflows/kb.json

Article frontmatter: title, type (how-to, troubleshooting, faq, or reference), owner, status, tags, and review_interval are required.
Workflow: draft → review → verified → stale → archived. Stale articles can go back to review.
Freshness: the janitor flags pages older than stale_days = 90.

wiki — team wiki

A team wiki with onboarding, decisions, and processes. Uses the default config.

knowledge

SCHEMA.md

playbook.md

index.md

welcome.md

architecture.md

how-we-work.md

decisions

onboarding

processes

reference

.kiwi

processes/ ships with deployment, dev setup, and incident response pages. reference/ has an FAQ and a glossary.

memory — agent memory

Episodic and semantic memory for agents. Agents write short-lived observations to episodes/ and consolidate them into durable pages in pages/. Uses the default config.

knowledge

SCHEMA.md

playbook.md

index.md

log.md

episodes

pages

.kiwi

Use kiwi_remember to write an episode, kiwi_memory_report to find episodes not yet consolidated, and kiwi_forget to mark a memory as superseded. See Episodic memory.

tasks — task tracking

Tasks as markdown files, ready for the kanban board and claims. Uses the default config.

knowledge

SCHEMA.md

WORKFLOW.md

playbook.md

index.md

tasks

.kiwi

schemas/task.json

workflows/tasks.json

Task frontmatter: type: task, title, status, and priority (an integer) are required. Optional fields include category, effort, assignee, parent, blocked-by, due, and claim fields such as claimed-by and lease-expires. Workflow actions read workflow and state, which is what the shipped example pages use.
Workflow: backlog → todo → in_progress → review → done, with cancelled reachable from every open state. in_progress has a WIP limit of 5 and review a limit of 3.

tasks/example-task.md

---
type: task
title: Implement auth flow
workflow: tasks
state: in_progress
priority: 1
assignee: agent:eng-bot
blocked_by: []
---

Advance a task through the workflow:

Terminal

curl -X POST 'http://localhost:3333/api/kiwi/workflow/advance' \
  -H 'Content-Type: application/json' \
  -d '{"path":"tasks/example-task.md","target_state":"review","actor":"agent:bot"}'

data — structured records

Collections of structured records with DQL dashboards.

knowledge

SCHEMA.md

playbook.md

index.md

collections

dashboards

imports

.kiwi

config.toml

schemas/record.json

Records need title and type. Dashboards in dashboards/ combine kiwi-query and kiwi-chart blocks. Fill collections with kiwifs import. See DQL.

cms — headless CMS

A git-native CMS with an editorial workflow, a public reader at /p/*, and feeds.

knowledge

SCHEMA.md

playbook.md

index.md

authors

blog

docs

pages

.kiwi

config.toml

schemas/blog-post.json

workflows/editorial.json

Post frontmatter: title, slug, type: blog-post, author, category, tags, published, and published_at are required.
Workflow: draft → review → scheduled → published → archived. Review can also publish directly.
Feeds: GET /api/kiwi/feed.xml (Atom) and GET /api/kiwi/feed.json (JSON Feed).

See Publishing.

runbook — operations

Runbooks with severity, affected services, and execution tracking.

knowledge

SCHEMA.md

playbook.md

index.md

example-high-cpu.md

.kiwi

config.toml

schemas/runbook.json

templates/runbook.md

Runbook frontmatter: type: runbook, title, trigger, severity (P1–P4), owner, and services are required. Track runs with last_executed, last_outcome, and execution_count.
Execution staleness: the janitor flags runbooks not executed in 90 days, and runbooks whose last_outcome is failure.
services is a typed link field, so services appear as edges in the graph.

adr — architecture decisions

Architecture Decision Records in MADR format.

knowledge

SCHEMA.md

playbook.md

index.md

decisions

.kiwi

config.toml

schemas/adr.json

templates/adr.md

workflows/adr.json

ADR frontmatter: type: adr, title, status, date, and deciders are required.
Workflow: proposed → accepted → deprecated → superseded.
Auto-numbering: new files in decisions/ get the next adr_number, configured with [format_hooks.auto_sequence].
supersedes, superseded_by, and contradicts are typed links, so the graph shows decision chains.

prompt — prompt registry

Versioned prompts with evaluation rubrics. Git history records every prompt change.

knowledge

SCHEMA.md

playbook.md

index.md

system-prompts

task-prompts

evaluation

.kiwi

config.toml

schemas/prompt.json

schemas/rubric.json

Prompts need type: prompt, title, model, and label (production or staging). Optional fields such as temperature, eval_score, success_rate, and variant_of make prompts queryable with DQL.

research — papers and notes

A research library with a reading pipeline and typed citation links.

knowledge

SCHEMA.md

playbook.md

index.md

papers

notes

reviews

.kiwi

config.toml

schemas/paper.json

workflows/reading.json

Paper frontmatter: type: paper, title, authors, year, venue, workflow, and state are required. doi, bibtex_key, and cites are optional.
Workflow: unread → reading → annotated → summarized → incorporated, with a WIP limit of 10 on reading.
cites, contradicts, extends, and reviews are typed links. The kiwi_cite MCP tool creates a paper page from a DOI or arXiv ID.

log — append-only events

An append-only event log with one file per day.

knowledge

SCHEMA.md

playbook.md

index.md

events

.kiwi

config.toml

schemas/event.json

Each daily file sets append_only: true in its frontmatter, and KiwiFS rejects overwrites of such files. Add entries with POST /api/kiwi/file/append or the kiwi_append MCP tool. Each entry is an H2 heading such as ## 2026-06-20T09:00:00Z | system.startup.v1, and git history provides tamper evidence.

blank — empty

No sample pages, no SCHEMA.md, and no playbook. You get the files listed in What every template includes and nothing else.

List templates over HTTP

A running server lists its templates at /api/init-templates, outside the /api/kiwi prefix:

Terminal

curl http://localhost:3333/api/init-templates

Response

{
  "templates": [
    {
      "id": "adr",
      "label": "Architecture Decision Records",
      "description": "Architecture Decision Records with MADR format, status workflow, and JSON Schema"
    },
    {
      "id": "blank",
      "label": "Blank",
      "description": "Empty workspace with Kiwi config only"
    }
  ]
}

Templates are sorted by label, with blank last.

Quickstart

Install and run KiwiFS.

Agent playbook

How agents use SCHEMA.md and playbook.md.

Workflows

State machines behind kanban boards and editorial pipelines.

Schemas

Validate frontmatter on write.

Getting Started

Core Concepts

Widgets

Guides

Data Import

Data Export

Deployment

Available templates

What every template includes

Export agent rules

Template reference

List templates over HTTP

Quickstart

Agent playbook

Workflows

Schemas

​Available templates

​What every template includes

​Export agent rules

​Template reference

​List templates over HTTP

​Related

Quickstart

Agent playbook

Workflows

Schemas

Available templates

What every template includes

Export agent rules

Template reference

List templates over HTTP

Related