AutoGineer
Generated banner for ThinkWiki
Generated logo for ThinkWiki

ThinkWiki

LLM wiki kit

First committed by wzdavid

Published in wzdavid/ThinkWiki. Publisher page

38 GitHub stars

What it is

Install once, talk to your agent. ThinkWiki turns documents and notes into a local Markdown wiki with inbox review, content knowledge graph, and browsable HTML workspace.
— the description by wzdavid

AutoGineer overview, from the brain’s own files

ThinkWiki is an agent-native skill designed to turn scattered information into a structured, durable local workspace. By integrating directly with skills-compatible AI agents, it allows users to manage their knowledge base through natural conversation rather than manual CLI commands. It supports a wide range of input formats, including PDFs, web pages, and various document types, organizing them into a system that features an inbox for review, a local viewer, and an interactive content knowledge graph. The system is built to be local-first, ensuring that Markdown files remain the source of truth, while providing HTML-based interfaces for visualizing relationships between topics, entities, and claims.

How it works

Memory model: LLM wiki

AutoGineer overview, from the brain’s own files

ThinkWiki functions as a skill installed into an AI agent's environment. It uses a set of Python scripts to manage a local directory of Markdown files. The system processes inputs through an inbox workflow, where content is reviewed before being ingested into the wiki. It generates a series of HTML outputs, including a workspace home, an inbox console, a page viewer, and a graph explorer. Users interact with these features by asking their agent to perform tasks like clipping content, building the knowledge graph, or serving the workspace via a local loopback server.

Folders
FolderRoleFiles
.githubagent-config1
docsdocs41
scriptsscripts34
templatestemplates13
teststests1

The agent reads

Requirements

Runtimes: Python

API keys (by name): THINKWIKI_EMBED_API_KEY, THINKWIKI_LLM_API_KEY

Set it up

  1. Clone the repository at the commit AutoGineer scanned:

    git clone https://github.com/wzdavid/ThinkWiki.git && cd ThinkWiki && git checkout 89f477d7a4095ba2ecf49ef9332e240c6f1208a1

  2. cd ThinkWiki && claude

  3. AutoGineer scanned these files at commit 89f477d.

  4. Not scanned by AutoGineerThe author’s README also describes installing it from a clone of its repository. Read its install section: AutoGineer did not scan that path. https://github.com/wzdavid/ThinkWiki/blob/89f477d7a4095ba2ecf49ef9332e240c6f1208a1/README.md#manual-install-optional

  5. First committed by wzdavid. Published in wzdavid/ThinkWiki. Licence: MIT License.

Usage

AutoGineer overview, from the brain’s own files

  • Initialize a new local wiki workspace
  • Clip web pages, files, or text into an inbox for later processing
  • Ingest reviewed items into the permanent wiki structure
  • Query the wiki for information with evidence-based responses
  • Generate and visualize a content knowledge graph
  • Serve the wiki workspace as a local HTML site for browser-based navigation
“Create a local knowledge base named Research Notes in my workspace.”
The agent initializes a new ThinkWiki workspace structure in the specified directory.
“Clip this article into the inbox: https://example.com/article”
The agent saves the content of the URL into the ThinkWiki inbox for later review.
“What did we already decide about context budgets?”
The agent searches the existing wiki pages and provides an answer supported by evidence from the knowledge base.
“Build the knowledge graph and open the workspace home in my browser.”
The agent generates the graph visualization and starts the local server to display the workspace in the browser.

Documentation by wzdavid

ThinkWiki

Image: Release Image: License Image: Python

ThinkWiki is an agent-native local knowledge base skill. Talk to your agent to turn scattered documents, web pages, notes, and conversations into a durable Markdown workspace — with inbox review, local browsing, and an interactive content knowledge graph you can open in a browser.

Chinese version: README.zh.md

Repository: https://github.com/wzdavid/ThinkWiki

Works With Skills-Compatible Agents

ThinkWiki follows the open Agent Skills format (SKILL.md + bundled scripts). If your agent can install skills and run local commands, it can run ThinkWiki.

You do not need to memorize CLI commands. Install the skill once, then create and manage your wiki entirely through conversation.

Agent / HostTypical use
Claude CodeInstall the repo as a skill, then ask it to build and query your wiki
OpenClawFull skill workflow; use serve + the built-in browser to open HTML outputs
TraeInstall under .trae/skills, then manage the wiki in chat
Hermes AgentLocal skill directory + shell execution
OpenAI Codex / Codex CLISkills-compatible coding agents with local file access
CursorAgent mode with project skills
Gemini CLISkills-compatible CLI agent
GitHub CopilotAgent / coding workflows in VS Code that support skills
Other skills hostsAny product that loads SKILL.md and can run python3 scripts/thinkwiki ... locally

Requirement: a local environment with Python 3. ThinkWiki bootstraps its own runtime on first use. Pure cloud chat without local file or shell access cannot run a local wiki.

Get Started: Install Through Conversation

Point your agent at this repository and ask it to install the skill for you.

Example prompt:

Please install the ThinkWiki skill from https://github.com/wzdavid/ThinkWiki , bootstrap the runtime, and run a health check to confirm everything works on my machine.

The agent will typically clone or copy the skill into your host's skills directory, run bootstrap, and verify capabilities with doctor. You should not need to run those steps yourself unless you prefer manual setup.

After installation, start using it immediately:

Create a local knowledge base named My Wiki in my workspace.

Clip this article into the inbox first: https://example.com/article

Import this PDF into the wiki: /path/to/file.pdf

Answer from my wiki: what did we already decide about context budgets?

Build the knowledge graph and open the workspace home in my browser.

The agent reads SKILL.md, maps your intent to stable ThinkWiki actions, and reports what changed — including paths to Markdown pages and HTML outputs.

Why ThinkWiki

  • Agent-first: users talk to an agent, not to a long list of commands.

  • Local-first: Markdown files remain the source of truth on your machine.

  • HTML-first: inbox, viewer, graph, and governance pages are real browsable workspaces.

  • Knowledge-first: the graph is a content knowledge graph — topics, concepts, entities, claims, and governance — not just a file-link map.

What It Produces

ThinkWiki generates a workspace that includes:

  • output/index.html: unified workspace home

  • output/inbox/index.html: inbox review console

  • output/viewer/index.html: local page browser

  • output/graph/index.html: interactive graph explorer

  • output/graph/report.html: graph governance report

  • output/graph/entity-merge-review.html: entity merge review page

  • output/graph/entity-merge-plan.html: entity merge dry-run preview when requested

Preview
Workspace Home (output/index.html)

The unified workspace home summarizes recent changes, recommended next actions, inbox backlog, and a graph snapshot so you can decide whether to read, ingest, or explore.

Image: ThinkWiki workspace home preview

Inbox Review (output/inbox/index.html)

Review clipped web pages, files, and notes before formal ingest. Items are grouped into ready, review, and weak so you can process the highest-value captures first.

Image: ThinkWiki inbox review preview

Local Viewer (output/viewer/index.html)

Browse the entire wiki by page type, confidence, and status. Open any page directly in the viewer without leaving the local HTML workspace.

Image: ThinkWiki local viewer preview

Knowledge Graph (output/graph/index.html)

Explore the content knowledge graph interactively. Switch between knowledge, document, and suggested views to inspect semantic relations and candidate edges.

Image: ThinkWiki knowledge graph preview

Graph Governance Report (output/graph/report.html)

Inspect graph health signals such as isolated pages, hub stubs, fragile bridges, suggested links, and entity merge candidates before making structural changes.

Image: ThinkWiki graph governance report preview

Everyday Commands (In Plain Language)

Once ThinkWiki is installed, these are the kinds of requests users make most often:

You sayThinkWiki does
Create a wiki called Research NotesInitializes a local workspace
Clip / save this URL or file to inboxCaptures content for later review
Ingest the ready inbox itemsPromotes reviewed captures into wiki pages
What does my wiki say about X?Evidence-first Q&A from existing pages
Save this answer as a query pagePersists high-value outputs
Show me the knowledge graphBuilds/refreshes graph HTML
Review entity merge candidatesSurfaces alias collisions for manual confirmation
Open the wiki workspace in my browserRuns serve and returns http://127.0.0.1:8765/index.html

Core Capabilities

  • Initialize a local wiki workspace

  • Clip web pages, files, or pasted text into an inbox for later review

  • Import Markdown, PDF, DOCX, XLSX, XLS, PPTX, web pages, and plain text

  • Answer questions from existing wiki pages with evidence-first behavior

  • Save high-value outputs as query, synthesis, decision, or concept pages

  • Build a content knowledge graph with knowledge, document, and suggested views

  • Generate entity governance outputs, including ambiguous alias review and deterministic merge flows

  • Check runtime health, workspace status, and graph governance state

Content Knowledge Graph

ThinkWiki v1.6.0 builds a schema v2 graph with default_view = knowledge.

The knowledge view can contain:

  • page-backed nodes such as source, topic, concept, decision, synthesis, query, and entity

  • claim nodes extracted from structured wiki content

  • semantic relations such as about, belongs_to, depends_on, asserts, supports, contradicts, and suggests_related_to

This means the graph is not limited to raw file references. It can represent knowledge structure, evidence structure, and entity governance directly.

Browse HTML Outputs

Agent chat UIs usually cannot render ThinkWiki HTML inline. Ask your agent to start the built-in loopback server when you want to inspect inbox, viewer, graph, or governance pages:

Start the ThinkWiki output server and give me the workspace URL.

Or run it yourself:

python3 scripts/thinkwiki serve --root /path/to/my-wiki

By default this serves <wiki-root>/output/ at http://127.0.0.1:8765/. Start from http://127.0.0.1:8765/index.html.

On OpenClaw, after serve is running, the agent can also open the workspace home with:

openclaw browser --browser-profile openclaw open http://127.0.0.1:8765/index.html

On other agents, open the same URL in your system browser, or use whatever browser tool the host provides.

Use serve --print-urls when you only need the URL list without starting the server.

Environment Variables

Remote AI features are disabled by default. Configure only what you need:

Content generation (crystallize, digest)

Requires all three variables. Any OpenAI-compatible chat completion endpoint works (MiniMax, OpenAI, DeepSeek, local Ollama, etc.).

VariableRequiredPurpose
THINKWIKI_LLM_API_KEYYes, to enableAPI key for the configured LLM provider
THINKWIKI_LLM_BASE_URLYes, to enableChat completions URL, e.g. https://api.openai.com/v1/chat/completions
THINKWIKI_LLM_MODELYes, to enableModel name, e.g. gpt-4o-mini or MiniMax-M3
THINKWIKI_LLM_TEMPERATURENoOptional temperature override

Without a complete LLM configuration, crystallize and digest use local heuristics only.

Entity embedding (entity-merge-review, graph-report, health)

Enabled when THINKWIKI_EMBED_API_KEY is set. Defaults to SiliconFlow BGE-M3; override the endpoint or model if needed.

VariableRequiredPurpose
THINKWIKI_EMBED_API_KEYYes, to enableAPI key for the embedding provider
THINKWIKI_EMBED_BASE_URLNoEmbedding URL(s), comma-separated; default https://api.siliconflow.cn/v1/embeddings
THINKWIKI_EMBED_MODELNoEmbedding model; default BAAI/bge-m3

Register a free SiliconFlow key at https://siliconflow.cn if you want the default embedding setup.

Manual Install (Optional)

If you prefer to set up the skill yourself instead of asking an agent:

git clone https://github.com/wzdavid/ThinkWiki ThinkWiki
cd ThinkWiki
python3 scripts/thinkwiki bootstrap
python3 scripts/thinkwiki doctor --repo-root .

Then install the folder into your agent's skills directory according to that host's documentation.

CLI Reference (For Agents And Power Users)

ThinkWiki keeps one stable entry point that agents call under the hood:

python3 scripts/thinkwiki <command> [args]

Common commands: init, clip, ingest, inbox, viewer, graph, graph-report, entity-merge-review, entity-merge-apply, serve, health, status, doctor.

See SKILL.md for the agent behavior contract and intent mapping.

Repository Docs

  • README.md: concise English project overview

  • README.zh.md: Chinese overview

  • SKILL.md: skill behavior contract for agent hosts

  • CHANGELOG.md: release history

License

MIT

CHANGELOG.md

Changelog

v1.7.3

Changed

  • Removed legacy env var compatibility (MINIMAX_*, SILICONFLOW_API_KEY, BGE_ENDPOINTS). Only THINKWIKI_LLM_* and THINKWIKI_EMBED_* are accepted.

v1.7.2

Security

  • Added SSRF protections for user-supplied URL fetches in ingest.py and clip.py via url_safety.py (scheme allowlist, private/metadata IP blocking, redirect validation). Set THINKWIKI_ALLOW_PRIVATE_URL_FETCH=1 only for local testing if you intentionally need loopback fetches.

  • serve now refuses non-loopback --host values unless --allow-lan is passed explicitly.

  • Upgraded cryptography from 48.0.0 to 48.0.1 (GHSA-537c-gmf6-5ccf).

Changed

  • Replaced MiniMax-specific m27_client.py with provider-neutral llm_client.py using OpenAI-compatible chat completions. LLM features stay disabled unless THINKWIKI_LLM_API_KEY, THINKWIKI_LLM_BASE_URL, and THINKWIKI_LLM_MODEL are all set.

  • Replaced bge_client.py with embed_client.py and unified env vars: THINKWIKI_EMBED_API_KEY (required to enable), THINKWIKI_EMBED_BASE_URL (defaults to SiliconFlow), THINKWIKI_EMBED_MODEL (defaults to BAAI/bge-m3).

  • Added ai_config.py as the single source of truth for optional remote AI configuration.

  • crystallize, digest, and entity merge now print explicit notices before sending content to configured remote APIs.

  • Deprecated legacy env vars (MINIMAX_*, SILICONFLOW_API_KEY, BGE_ENDPOINTS) with stderr warnings; they remained supported until v1.7.3.

v1.7.1

Security

  • Removed unauthorized hardcoded BGE embedding endpoints and default API key from bge_client.py. v1.7.0 had shipped with private HTTP endpoints and a baked-in shared key that did not match the documented SILICONFLOW_API_KEY flow. Embedding now uses SiliconFlow (https://api.siliconflow.cn/v1/embeddings) and requires an explicit SILICONFLOW_API_KEY; there is no default key and no silent third-party fallback.

  • bge_embed now fails fast when SILICONFLOW_API_KEY is unset, matching utils.py and the README/SKILL documentation.

v1.7.0

Added

  • Added scripts/m27_client.py — MiniMax M2.7 HTTP client with retry, backoff, 4xx short-circuit, and degraded fallback, shared by crystallize and digest.

  • Added scripts/bge_client.py — BGE-M3 embedding client (SiliconFlow BAAI/bge-m3, OpenAI-style /v1/embeddings) for semantic entity matching, with multi-endpoint failover and explicit auth-error handling.

  • Added BGE-M3 embedding branch to ambiguous_entity_merge_candidates in utils.py so entity alias groups can be detected by semantic similarity in addition to string matching. Configured via SILICONFLOW_API_KEY (optional; degrades gracefully).

  • Added MINIMAX_API_KEY and SILICONFLOW_API_KEY environment variable documentation across SKILL.md, README.md, and README.zh.md.

Changed

  • Replaced ~500 lines of heuristic content generation in crystallize.py with m27_crystallize() calls, and ~350 lines in digest.py with m27_digest() calls, removing dead constants and helper functions.

  • Updated _post_json in bge_client.py to accept an explicit api_key parameter, send User-Agent: ThinkWiki/1.0, and short-circuit on 401/403 auth failures instead of retrying.

  • Updated _resolve_api_key call site to the bge_embed entry point so key resolution happens once per batch, not per request.

  • Updated bge_embed docstring with worst-case latency formula and 4xx short-circuit behavior.

  • Updated m27_crystallize and m27_digest fallback messages to defend against empty titles.

Fixed

  • Fixed bge_client 401/403 handling: auth failures now raise BgeServiceUnavailable immediately instead of silently cycling through endpoints.

  • Fixed m27_client fallback stderr warning to handle None title gracefully.

  • Fixed m27_client _parse_result to strip thinking tags before JSON parsing, preventing unnecessary heuristic fallback on reasoning model output.

  • Fixed utils.py BGE embedding branch: exceptions now log warnings to stderr instead of silently passing.

v1.6.0

Added

  • Added a schema v2 content knowledge graph with default_view = knowledge, including page-backed nodes, extracted claim nodes, and semantic relations such as about, belongs_to, depends_on, asserts, supports, contradicts, and suggests_related_to.

  • Added entity governance to the knowledge graph so page-backed entity nodes now carry aliases, can surface ambiguous identity collisions, and participate in deterministic merge review workflows.

  • Added an entity-merge-review command that writes output/graph/entity-merge-review.{json,md,html} for manual review of ambiguous entity alias groups.

  • Added entity-merge-apply so users can canonicalize one entity page, convert the rest into merged stubs, and rebuild downstream viewer / graph / governance outputs in one deterministic step.

  • Added entity-merge-apply --dry-run so users can generate output/graph/entity-merge-plan.{json,md,html} before writing any entity pages.

  • Added a serve command that exposes <wiki-root>/output/ over loopback HTTP (default http://127.0.0.1:8765/) so agent hosts can browse inbox, viewer, graph, and governance pages in a browser.

  • Added README.zh.md as the Chinese project overview alongside the English README.

Changed

  • Updated HTML workbench outputs (output/index.html, inbox, viewer, graph, and governance pages) to English so repository-facing previews and agent-host browsing stay consistent.

  • Refreshed demo wiki outputs and README screenshot assets to match the current English HTML workspace.

  • Rewrote README.md and SKILL.md around Agent Skills installation and conversational usage, including supported agent hosts and the recommended serve workflow for browsing HTML outputs.

  • Updated graph-report, status, health, and output/index.html so entity counts, alias counts, ambiguous alias groups, and ambiguous entity counts are surfaced across terminal summaries and HTML workbench outputs.

  • Updated graph construction and ingest so merged entity stubs continue to resolve old titles and aliases to the canonical entity page without polluting the active knowledge graph.

  • Updated the output home so entity-merge-review.html and entity-merge-plan.html are treated as first-class governance artifacts alongside the graph report.

v1.5.1

Changed

  • Updated graph-report terminal output to stay ASCII-safe across platforms, preventing Windows CI failures caused by console encoding when the report summary contains non-ASCII text.

v1.5.0

Added

  • Added a deterministic health command so users can check workspace structure, inbox consistency, and stale outputs without invoking heavier content workflows.

  • Added a compact status command so users can quickly inspect page counts, inbox readiness, and the current viewer/graph/inbox output state from the terminal.

  • Added a batch-clip command so users can collect a whole directory or a manifest of source / url / text items into the inbox in one run.

  • Added a batch-ingest command so users can promote ready inbox items into the wiki in batches, with --dry-run, --limit, and quality-based filtering.

  • Added a deterministic graph-report command so users can turn the current graph into a governance report with isolated pages, hub stubs, fragile bridges, isolated clusters, and top graph fixes.

Changed

  • Updated output/inbox/index.html so the Ready To Ingest section now recommends batch-ingest dry-run and execution commands before the per-item ingest commands.

  • Updated status, health, and output/index.html so the new graph governance report is surfaced in terminal summaries and the HTML workspace home.

  • Updated graph-report to also write output/graph/report.html, making the governance report directly browsable like the rest of ThinkWiki outputs.

  • Updated demo generation and committed demo outputs so repository-facing pages no longer expose local absolute paths, and removed internal competitive research notes from docs/.

v1.4.0

Added

  • Added structured sidecar metadata for web clips in normalized/inbox/*.json, including adapter, site name, author, publish date, and source URL.

  • Added explicit clip --adapter auto|wechat|generic selection so the inbox capture flow can evolve toward adapter-based web extraction.

  • Added inbox extraction quality states (ready, review, weak) so the review page can highlight which clips are safe to ingest and which still need manual checking.

  • Added clip --mode auto|wait so webpage capture can retry for a short window before writing inbox artifacts, and record whether the wait completed or timed out.

  • Added clip --media ask|always|never so webpage images can stay remote, be marked for later review, or be localized into normalized/assets/inbox/... during capture.

  • Added structured capture reasons such as loading_placeholder, body_too_short, and metadata_sparse, so inbox review can explain why a web clip still needs attention.

Changed

  • Updated output/inbox/index.html and the workspace inbox cards so clipped webpages now surface adapter, source, author, publish date, and metadata links during review.

  • Updated the WeChat extraction path so common embedded code blocks are normalized before Markdown conversion, making technical articles easier to preserve.

  • Updated output/inbox/index.html into a grouped review console with Ready To Ingest, Needs Review, and Weak Captures sections, plus priority commands for the next ingest steps.

  • Updated output/index.html so the workspace home now highlights ready inbox items first and links directly to the ready review section.

v1.3.0

Added

  • Added a new clip command and inbox storage flow so webpages, pasted text, and local files can be collected first and ingested into the wiki later.

  • Added an output/inbox/index.html review page so users can browse pending inbox items and copy the next ingest command without leaving the browser.

Changed

  • Updated output/index.html so the workspace home now shows an Inbox Queue, inbox counts, clip-driven next actions, and a direct entry to the inbox review page.

  • Updated clip so it backfills missing raw/inbox and normalized/inbox directories for older wikis, regenerates output/inbox/index.html, and refreshes the output home after each capture.

v1.2.0

Added

  • Added a workspace-style output/index.html home that surfaces What Changed, Next Actions, Needs Attention, Graph Snapshot, and Featured Pages.

  • Added homepage recommendations that reuse graph insight data so users can move from the output hub into the right next action faster.

v1.1.0

Added

  • Added Graph Insights to output/graph/index.html so the graph page now surfaces key pages, bridge pages, weakly connected pages, and suggested links.

  • Added structured graph insight data to output/graph/graph.json and output/graph/graph.md for downstream analysis and richer summaries.

  • Added graph-side interactions for insight-driven exploration, including clickable insight cards and suggested-link highlighting in the SVG graph.

Changed

  • Updated the graph explorer README copy and demo screenshot so the new insight panel is visible in the repository front page.

v1.0.1

Fixed

  • Fixed GitHub Actions regression tests so spawned test scripts now prefer the repository runtime in .venv after bootstrap, instead of falling back to the system Python.

  • Fixed CI failures in ingest.py regression coverage caused by running document and directory ingest tests outside the bootstrapped ThinkWiki runtime.

v1.0.0

Released as ThinkWiki

  • First standalone release under the ThinkWiki brand.

  • Removed the legacy scripts/llm-wiki compatibility entry and standardized on scripts/thinkwiki.

  • Standardized runtime environment variables on THINKWIKI_*.

  • Renamed the regression test suite to tests/test_thinkwiki.py.

  • Refreshed demo outputs, page titles, and repository-facing docs to consistently use ThinkWiki.

Added

  • Added an offline graph viewer at output/graph/index.html so graph generation now produces a directly browsable HTML artifact.

  • Added richer graph node metadata including summary, confidence, status, updated date, path, and sources.

  • Added a shared output/index.html hub so users can open the graph page and viewer page from one place.

  • Added repository-hosted PNG screenshots for the viewer page and graph page so README can show real product output immediately.

  • Added a repository-hosted PNG screenshot for output/index.html so README now shows the entry hub alongside viewer and graph pages.

  • Added a reusable scripts/build_demo_wiki.py script to regenerate the demo wiki behind the README screenshots.

  • Added release notes tracking for ThinkWiki.

Changed

  • Updated README.md to better explain the single-skill installation model, quick start flow, and viewer/graph outputs.

  • Updated SKILL.md to treat graph and viewer tasks as HTML-first deliverables instead of raw data outputs.

  • Updated graph layout generation to use a denser, degree-aware column layout and auto-center selected nodes in the graph stage.

  • Updated graph edge rendering so references, links_to, includes, and cites now use distinct colors and line styles.

  • Updated graph exploration so users can switch between 1-hop, 2-hop, and full-graph visibility around the selected node.

  • Updated graph exploration so users can toggle individual edge types on and off when reading dense relationship maps.

  • Updated graph exploration so users can jump to all, concept, decision, and source views with one-click focus presets.

  • Updated graph node details so the side panel now summarizes per-edge-type relationship counts for the selected node.

  • Updated graph and viewer pages so they can jump to each other through URL hashes and printed file URIs.

  • Updated ask, query, digest, and crystallize so they mention the shared output/index.html hub whenever viewer or graph outputs already exist.

  • Updated output/index.html so it now acts like a lightweight product homepage with wiki title, generated date, page count, node count, edge count, recent pages, featured concept/decision pages, and a recommended next step.

Fixed

  • Kept wiki-backed graph nodes from losing their richer metadata when placeholder nodes are merged during graph construction.

98 files · 5 not served here

Advantages

  • Enables natural language management of a knowledge base without requiring the user to memorize complex CLI commands. evidence: README.md
  • Maintains Markdown files as the local source of truth, ensuring data portability and ownership. evidence: README.md
  • Provides a rich, interactive content knowledge graph that maps semantic relations like claims and entities rather than just file links. evidence: README.md

Limitations

  • Requires a local environment with Python 3 and cannot function in pure cloud-based chat environments without local file or shell access. evidence: README.md
  • Agent chat UIs typically cannot render the generated HTML interfaces inline, requiring the use of a separate browser or a local loopback server. evidence: README.md

Security

Risk level: No risk found

This brain runs

Agent instructions that run or fetch something (5)

  • SKILL.md: If the user wants a new workspace, run `init`.
  • SKILL.md: After `viewer`, `graph`, `inbox`, or `graph-report`, run `serve` when the user wants to browse the results.
  • SKILL.md: If a long-running `serve` process is not practical in the current session, run `serve --print-urls` and tell the user to start `serve` locally, then open the printed workspace URL.
  • SKILL.md: After starting `serve`, give the user the workspace home URL first (`http://127.0.0.1:8765/index.html` by default) and, in OpenClaw, offer to open it with `openclaw browser --browser-profile openclaw open <url>`. · uses the network
  • SKILL.md: If the host cannot keep a long-running process alive, run `serve --print-urls` to show the URLs and ask the user to start `serve` in a local terminal or background job.

Attribution and licence

First committed by
GitHub avatar of wzdavidwzdavid
Publisher
Wei Zhang (Community)
Source
wzdavid/ThinkWiki · commit 89f477d
Licence
MIT License (files served here)
  • ATTRIBUTION.md (written by AutoGineer)
Synced
Contributors
wzdavid

AutoGineer is not affiliated with or endorsed by wzdavid.

Install from source

5 of its files are not served here, so there is no ZIP. wzdavid/ThinkWiki at 89f477d