Wiki Agent#

A personal knowledge-base agent: it pours sessions and notes into a template-driven HTML wiki. Pages are real HTML files — open them in any browser, host them statically, version them with git. The agent only fills named slots in fixed templates and never hand-writes HTML / CSS. It ships with full-text search, automatic folder indexing (each folder's README.html is regenerated from its current contents), and a curating ingest pipeline.

Install#

openprogram programs install wiki

Dependencies are minimal: Jinja2 + PyYAML. The template shell uses Jinja2 + Bootstrap 5 (loaded from a CDN) — zero build steps, no Node required.

Usage#

The entry function is wiki_agent. In chat, describe what you want done to the wiki in natural language, for example:

• "ingest these notes about transformers" (pour the conversation above into the wiki)
• "enrich the Methodology landing page"
• "browse the vault"
• "check for broken links"
• "find pages about distillation"

Run it directly from the command line:

openprogram programs run wiki_agent -a task="Ingest the notes above into the wiki"

Function signature: wiki_agent(task, vault="", purpose="", audience="", ...). It is a dispatcher: using a next-step decision (decision.make) it routes the task to one of five internal operations — ingest / enrich / browse / lint / search — where each branch handler is plain Python calling wiki_agent_harness.Wiki. vault (the knowledge-base root), purpose, and audience are hidden parameters that do not appear in the chat tool form; when vault is not given it falls back to the runtime's working directory.

The harness itself stays deliberately loose: it provides only five things — the rendering layer, slot primitives, full-text search, automatic folder indexing, and the ingest pipeline. Everything else (moving / deleting / grepping / editing) the agent does with ordinary shell and file tools. A slot is a region of a page delimited by HTML comments; the agent edits one slot at a time, so repeated ingests accumulate cleanly without rewriting whole pages.

Dependency notes#

• Works without openprogram: with OpenProgram not installed, the pure-Python Wiki class and the CLI still work as usual — you just lose the wiki_agent chat entry point.
• Downstream projects (paper surveys, memory stores, CRMs, etc.) specialize the ingest pipeline by supplying their own templates and prompts — no fork needed.

Source and README: openprogram/functions/agentics/Wiki-Agent-Harness/, upstream repository Fzkuji/Wiki-Agent-Harness.