How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

mykb - Skill - OpenSmartRoute

Imported from gemquota/cosmos (components/mykb/AGENTS.md). Install upstream with npx skills add gemquota/cosmos --skill mykb. Copyright stays with the author.

AGENTS.md

This vault is a personal wiki knowledge iteration system for daily learning, formatted as an Open Knowledge Format (OKF) bundle. Use $okf for knowledge work.

OKF Bundle

This project is registered as an OKF bundle. The okf CLI is installed and available. Main verbs:

okf validate <dir> — check OKF v0.1 conformance
okf lint <dir> — curation-quality report
okf search <dir> <term> — find concepts
okf produce — create new concepts
okf maintain — keep in sync
okf server <dir> — serve as interactive graph

This vault also follows the Obsidian Wiki System pattern. Use $obsidian-wiki-system for maintenance workflows, or $okf for OKF operations.

Core Layers

raw/: Original material (links, PDFs, screenshots, transcripts, clippings). Store here without rewriting source facts.
wiki/: Processed knowledge. Digested, linked, rewritten notes. Each file has YAML frontmatter with type, title, tags, timestamp.
templates/: OKF note templates (from the installed skill).
ops/: Operating layer — workflows, schema, prompts, maintenance rules.
daily/: Daily notes.

Working Principles

Read index.md (OKF bundle root) and wiki/index.md before editing.
Preserve the raw -> wiki -> index/log iteration loop.
One concept = one file. Links are the graph.
New pages must include YAML frontmatter with at minimum a type field.
Use [[wikilinks]] for internal vault pages alongside markdown links.
Update log.md after meaningful structural changes.
Run okf validate and okf lint after major edits.

Wiki Frontmatter (compatible with OKF)

---
type: concept    # Required by OKF §9.2
title: ""        # Display name
description: ""  # One-sentence summary
tags: []         # Cross-cutting labels
timestamp: ""    # ISO 8601

Extended frontmatter (obsidian wiki compatible):

---
status: seed      # seed | growing | stable | archive
source: []        # Links to source materials

Valid type values: source, concept, domain, project, question, synthesis, daily, index, log.

Common Actions

Ingest material: put original in raw/inbox/, then use $okf produce to create wiki concept pages.
Process source: use ops/prompts/ingest-source.md to create a wiki/sources/ page from inbox material.
Synthesize: after collecting related sources, create a synthesis page in wiki/syntheses/ that cross-links concepts.
Weekly review: update wiki/syntheses/weekly-review.md, merge duplicate notes, check isolated pages.
Query knowledge: use okf search from the CLI, or $okf search from the agent.
Health check: okf validate . for conformance, okf lint . for curation.
Serve graph: okf server . to browse as an interactive knowledge graph.

Web Dashboard

The wiki is served via server.py (Python HTTP server) on port 8826.

# Start dashboard
bash start.sh
# Or from anywhere (after sourcing ~/.bashrc):
mykb

Dashboard URL: http://127.0.0.1:8826

API Endpoints

Endpoint	Method	Purpose
`/`	GET	Serves `index.html` (SPA viewer)
`/files.json`	GET	Enriched index of all `.md` files: `{path, type, title, tags}`
`/api/file?path=<rel>`	GET	Serves raw markdown content
`/api/stats`	GET	System statistics
`/api/v2/search/hybrid?q=`	GET	Hybrid search (BM25 + TF-IDF)
`/api/v2/search/build`	GET	Rebuild search index
`/api/v2/graph/topology?root=&depth=`	GET	Subgraph filtering
`/api/v2/history/log/<path>`	GET	Git commit history for a file
`/api/v2/history/snapshot?path=&ts=`	GET	File content at a point in time
`/api/v2/health/lint`	GET	Wikilink integrity report
`/graph.json`	GET	Full knowledge graph (nodes + edges)
`/search?q=`	GET	Basic TF-IDF search

Viewer Features

Docs tab: Browse all markdown files in a sidebar, type-grouped or directory-grouped
Graph tab (Ctrl+G): Force-directed knowledge graph with topology filtering
Actions tab (Ctrl+H): System stats, linter, search index rebuild
Search (Ctrl+K): Full-text search across all documents
Theme toggle (T): Cycles light → dark → AMOLED ultra-black

Session Capture Hooks

Python hooks capture agent turns and buffer them for knowledge extraction.

Architecture

Agent turn → PostToolUse hook → buffer (.ndjson)
Session end → Stop hook → signal file

Components

hooks/post-tool-use.py — captures each turn to .wiki-daemon/buffers/<session>.ndjson
hooks/session-stop.py — writes session-end signal to .wiki-daemon/buffers/signals/

Usage

Hooks are configured in the Codex settings and run automatically. No manual start/stop required.

Active .wiki-daemon Scripts

These scripts are referenced by server.py and actively used:

search_fusion.py — hybrid search engine (BM25 + TF-IDF + RRF)
kb_linter.py — wikilink integrity, orphan detection
temporal_engine.py — Git-backed history, auto-commit, time-travel retrieval

Standalone utility scripts (run manually) are archived in .wiki-daemon/archive/unused-scripts/.

File Inventory

Top-level

File	Purpose
`server.py`	HTTP server with API endpoints
`index.html`	SPA viewer (Docs/Graph/Actions tabs)
`start.sh`	Server launcher (run from any directory)
`build-export.py`	Concatenate project into two markdown files
`build-index.py`	Build enriched index.json from wiki frontmatter
`AGENTS.md`	This file — agent instructions
`index.md`	OKF bundle root
`Home.md`	Daily knowledge entry point
`log.md`	Iteration log
`README.md`	Project overview

Backup/Archive

.wiki-daemon/archive/ — archived scripts, dead files, old viewers
mykb-code.md / mykb-content.md — full project export (generated by build-export.py)

Install mykb and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

mykb

About