v0.4.1Python library & command-line tool

WIKI-STUDIO

Give your agents
a living wiki.

Turn source notes into a curated Markdown knowledge base. Propose changes, review what will land, and give your agents knowledge they can read, cite, and maintain.

Get started View on GitHub

$python -m pip install wiki-studio

Python 3.11–3.13 · Standard library only.
Use it independently or alongside MemKernel’s wiki bridge.

knowledge/wiki/decisions/database.mddraft

A DECISION WITH ITS CONTEXT

Use PostgreSQL for orders

Keep concurrent inventory updates and order writes in one transaction.

Sourcepath: notes/database-decision.md

DecisionContextAlternativesConsequences

Illustrative page excerpt · Drafts stay reviewable.

KNOWLEDGE YOU CAN INSPECT

Plain MarkdownValidated proposalsCited sourcesApache-2.0

01FROM SOURCE TO SHARED KNOWLEDGE

A wiki that grows
with your project.

Keep decisions, concepts, workflows, and incident lessons in typed pages, connected by an index and an edit log. Every stage leaves files you can inspect.

01 / PROPOSE

Bring in the source.

Ingest a note through a copilot CLI, or propose pages that you or your agent wrote. The result is a plan and page previews.

Reviewable outputplan.json + preview/*.md

02 / PROMOTE

Review. Then apply.

Inspect the proposal before promotion. Wiki-Studio validates the staged pages and links before applying the changes.

Check before writingwiki-studio promote … --dry-run

03 / MAINTAIN

Ask, find, and improve.

Query the wiki with citations, find pages without a model, and lint the structure as your project’s knowledge changes.

Ongoing healthwiki-studio lint --health

The structural gate checks page shape, placement, links, and reachability. Reviewing the facts and the quality of the prose remains part of curation.

02YOUR FIRST WIKI

Start with one project.
Keep what you learn.

Install the package in a Python environment, initialize your wiki, and inspect its layout. Add a copilot backend when you want model-assisted ingestion and answers.

1. Install and initialize

From your project directory on macOS or Linux, using an installed Python 3.11, 3.12, or 3.13:

python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install wiki-studio

wiki-studio --version
wiki-studio init --title "My project knowledge"
wiki-studio info
wiki-studio lint

On Windows, use py -3.11 -m venv .venv and activate with .venv\Scripts\Activate.ps1 in PowerShell. init creates missing index and log files and preserves existing ones.

2. Ingest a real source

Choose an existing source file within your project. This example uses Claude Code, installed and authenticated separately.

# Replace the path with an existing source file.
wiki-studio ingest notes/database-decision.md --backend claude

# Inspect the printed proposal directory:
# plan.json describes edits; preview/ contains pages.
# Replace PROPOSAL_DIR with that directory.
wiki-studio promote "PROPOSAL_DIR" --dry-run
wiki-studio promote "PROPOSAL_DIR"

wiki-studio query "Why did we choose PostgreSQL?" --backend claude
wiki-studio lint --health

Promote after reviewing the proposal and a successful dry run. An ingest can reject a source or fail validation; use the reported rationale before proceeding.

Bring your copilot. Supported backends are claude, codex, cursor (the cursor-agent binary), and pi (the pi-code binary). Choose one with --backend or WIKI_STUDIO_BACKEND. Source text and selected pages are passed to that CLI. Initialization, direct proposals, promotion, page finding, and structural linting need no LLM.

Where are my files stored?

Default Wiki-Studio locations
Purpose	Default location
Project root	Git repository root, or the current directory outside Git
Curated wiki	`<root>/knowledge/wiki`
Pending proposals	`<root>/doc_internal/proposals`
Query log	`<root>/doc_internal/wiki-query-log.jsonl`
Ingest audit log	`<wiki-dir>/.audit/ingest-log.md`

Override locations with --root, --wiki-dir, and --proposals-dir, or their WIKI_STUDIO_* environment variables. Flags take precedence. Run wiki-studio info --json to inspect the resolved paths.

03A SMALL, COMPOSABLE TOOLSET

One workflow.
Nine commands.

Use the shell directly or let your agent call it. Each command has its own help: wiki-studio <command> --help.

Wiki-Studio commands and model requirements
Command	What it does	Model needed?
`init`	Create missing wiki index and log files.	No
`ingest <source>`	Evaluate a source and draft a proposal containing one or more page edits.	Yes
`propose --from <dir>`	Validate pages you supply and package them as a proposal.	No
`promote <proposal-dir>`	Validate and apply a proposal; supports a dry run.	No
`query "question"`	Load the index and selected pages, then produce an answer with citations.	Yes
`find "question"`	Show the pages a query would select, without generating an answer.	No
`lint`	Check page metadata, structure, links, reachability, and audit format.	Only for the optional contradiction check
`audit-flush`	Explicitly commit pending ingest audit entries to Git; supports a dry run.	No
`info`	Show active paths, their configuration sources, and the selected schema.	No

Find pages without a model

find and query start with the index, then rank linked pages by word overlap with the question. The default limit is five additional pages; set --max-pages to change it. This is keyword-based selection and can miss relevant pages.

Keep the knowledge healthy

lint --health adds stale source, orphan, and cross-link checks. Add --backend for contradiction checks and --strict to make audit and health findings fail the command.

Read the published evaluation and its limitations ↗

04BUILD IT INTO YOUR TOOLS

The same workflow.
A typed Python API.

Use wiki_studio.api in a harness, an automation, or your own application. The CLI calls the same functions.

01
One path configuration
WikiPaths defines the project root, wiki, and proposals directories.
02
Results your code can inspect
Typed results expose selected pages, applied edits, citations, and validation findings.
03
Compose or generate
Pass your own page edits to api.propose, or use api.ingest and api.query with a backend.

Full library examples

knowledge.py · no model required

from pathlib import Path
from wiki_studio import api
from wiki_studio.paths import WikiPaths

paths = WikiPaths.for_root(Path.cwd())
api.init(paths, title="Project knowledge")

# Search the pages already in your wiki.
result = api.find(
    "database decision", paths, max_pages=5
)
for page in result.pages:
    print(page.title, page.file)

# Inspect structural and health findings.
report = api.lint(paths, health=True)
print(report.passed, report.violations)
if report.health:
    print(report.health.findings)

A new wiki has no content pages yet. Add pages through ingest or propose, then promote them before searching.

05MEMORY MEETS CURATION

Two tools.
A shared knowledge base.

MemKernel retains and retrieves a copilot’s working context. Wiki-Studio maintains structured Markdown knowledge. Their shared page schema connects memory to a wiki people can review.

MEMKERNEL

Capture the useful context.

Export long-term memories as draft pages with source references.

WIKI-STUDIO + YOUR REVIEW

Curate durable knowledge.

Review and maintain decisions, concepts, and workflows in Markdown.

MEMKERNEL

Make it available to recall.

Import curated wiki pages back into long-term memory.

Point both tools at the same wiki

Run these commands from your project root. Call MemKernel’s wiki_export and wiki_import tools from your MCP client; use Wiki-Studio’s CLI to inspect and lint the shared pages.

For an MCP client, also set MEMKERNEL_WIKI_DIR to that same absolute path in the server configuration. The shell exports configure processes started from this shell.

MemKernel wiki bridge guide

export MEMKERNEL_WIKI_DIR="$PWD/knowledge/wiki"
export WIKI_STUDIO_WIKI_DIR="$MEMKERNEL_WIKI_DIR"

wiki-studio info
wiki-studio lint --health

Release compatibility: the published MemKernel 0.4.1 package uses the shared wiki schema; install Wiki-Studio separately for its CLI. MemKernel’s current development branch adds the optional memkernel[wiki] dependency and routes exports through Wiki-Studio’s proposal/promotion gate. That extra is not included in the published MemKernel 0.4.1 metadata. Check the MemKernel releases before using the gated integration.

Wiki-Studio also works on its own. You don’t need MemKernel to create, curate, or query a wiki.

06OPEN SOURCE, ALL THE WAY DOWN

Explore the implementation.

Install from PyPI, read the schema, and see how other tools use the same API.

Make the next answer easier to find.

Start with your project’s notes. Build knowledge you can carry forward.

Install Wiki-Studio Issues & feedback ↗

Give your agents
a living wiki.

Use PostgreSQL for orders

A wiki that grows
with your project.

Bring in the source.

Review. Then apply.

Ask, find, and improve.

Start with one project.
Keep what you learn.

1. Install and initialize

2. Ingest a real source

One workflow.
Nine commands.

Find pages without a model

Keep the knowledge healthy

The same workflow.
A typed Python API.

Two tools.
A shared knowledge base.

Capture the useful context.

Curate durable knowledge.

Make it available to recall.

Point both tools at the same wiki

Explore the implementation.

Wiki-Studio on PyPI

Read the documentation

Understand the page schema

Build an integration

Make the next answer easier to find.

Give your agentsa living wiki.

Use PostgreSQL for orders

A wiki that growswith your project.

Bring in the source.

Review. Then apply.

Ask, find, and improve.

Start with one project.Keep what you learn.

1. Install and initialize

2. Ingest a real source

One workflow.Nine commands.

Find pages without a model

Keep the knowledge healthy

The same workflow.A typed Python API.

Two tools.A shared knowledge base.

Capture the useful context.

Curate durable knowledge.

Make it available to recall.

Point both tools at the same wiki

Explore the implementation.

Wiki-Studio on PyPI

Read the documentation

Understand the page schema

Build an integration

Make the next answer easier to find.

Give your agents
a living wiki.

A wiki that grows
with your project.

Start with one project.
Keep what you learn.

One workflow.
Nine commands.

The same workflow.
A typed Python API.

Two tools.
A shared knowledge base.