kib Roadmap

What's built, what's next, and what's deferred.

Shipped (v0.4.x)

Core (v0.2.0)

kib init — vault creation with auto-detected LLM provider
kib ingest — web, file, PDF, YouTube, GitHub extractors with dedup
kib compile — LLM-powered compilation with incremental builds, INDEX.md, GRAPH.md
kib search — BM25 full-text search with stemming and index persistence
kib query — RAG query engine with article retrieval and citations
kib chat — interactive REPL with conversation history
kib lint — 5 rules (orphan, broken-link, stale, frontmatter, missing)
kib skill — skill system with loader, runner, 3 built-in skills
kib watch — file watcher + local HTTP server for browser extension
kib export — markdown and HTML static site export
kib status — vault health dashboard
kib config — get/set/list vault configuration

LLM & Providers (v0.2.0)

Anthropic, OpenAI, Ollama providers with lazy-loading
Interactive provider setup flow (select provider, enter API key, auto-save)
Credentials stored at ~/.config/kib/credentials, auto-loaded on startup
LLM response cache with TTL

Distribution (v0.2.0)

Published to npm as @kibhq/core and @kibhq/cli
Standalone binaries for macOS/Linux via bun build --compile
Release Please: auto-versioning, changelogs, npm publish, binary builds
CI: Biome lint + 412 tests on every push

MCP Server (v0.2.0)

kib serve — expose vault as MCP tools over stdio
kib init auto-configures MCP in Claude Code, Claude Desktop, and Cursor
kib mcp to re-configure MCP clients independently
8 tools: kib_status, kib_list, kib_read, kib_search, kib_query, kib_ingest, kib_compile, kib_lint
2 resources: wiki://index, wiki://graph

CLI Polish (v0.3.0)

--json flag working consistently across all commands
--verbose flag for debug output
--dry-run for ingest (show what would be ingested without writing)
kib compile --dry-run visual diff preview
kib lint --fix actually runs auto-fixes (recompile stale, create missing articles)
Colored diff output showing what compile changed
<100ms cold start for kib --help

Smarter Compilation (v0.4.0)

Token usage tracking per compile pass and per source
Auto-summarize large sources before sending to LLM (chunk into 8K-token windows)
Smart context selection: for large vaults, send INDEX.md + summaries, not full articles
Token budget config: compile.max_tokens_per_pass
Warning when a single source exceeds the model's context window
Article merging: detect when two sources produce articles about the same topic
Compile cache integration: skip LLM call when cache hit matches
Retry with adjusted prompt when LLM returns malformed output (max 2 retries)
kib compile --source <path> to recompile a specific source
Streaming compile output: show article titles as they're generated
Parallel compilation: compile independent sources concurrently

Multi-Model Support (v0.4.0)

Config-based model selection per operation: compile.model, query.model, lint.model
fast_model used for lightweight ops (skills with model: "fast")
default model used for heavy ops (compile, query)

Other (v0.4.x)

Fall back to ~/.kib when no local vault found
Surface auth errors in compile and improve MCP provider feedback
Improved compile error handling
Website launched

v0.5.0 — Image + Vision Support & Packaging

Distribution

Homebrew formula for brew install kib
Auto-generate a CLAUDE.md in vault root on kib init for agent discovery

Image Extractor

kib ingest photo.png — send to vision model, get markdown description
Store description in raw/images/{name}.md with metadata
Support PNG, JPG, JPEG, WebP, GIF, SVG, BMP, TIFF
Vision API integration for Anthropic (Claude) and OpenAI (GPT-4V)
Whiteboard photo → structured notes extraction
Diagram → markdown description + detected labels
Screenshot → extract text content via vision model

Image References in Articles

Compile step can reference images from raw/images/ in wiki articles
HTML export includes images with proper relative paths
Image gallery view in HTML export

v0.6.0 — Browser Extension

Chrome Extension (MV3)

packages/extension/ — Chrome Manifest V3 extension
"Send to KB" button on any webpage
Content script: readability-based article extraction
Background script: turndown conversion + POST to localhost:4747/ingest
Popup UI: status indicator (green = kib watch running), tag input, vault selector
Optional: highlight text on page → send selection to KB
Build pipeline: bun build for extension, separate from CLI

Firefox Extension

Port Chrome extension to Firefox (WebExtension API)
Publish to Firefox Add-ons

v0.7.0 — Advanced Search

Vector/Semantic Search

Optional embedding-based search alongside BM25
Embedding provider: OpenAI text-embedding-3-small, or local via Ollama
Hybrid scoring: combine BM25 + cosine similarity (Reciprocal Rank Fusion)
Store embeddings in .kb/cache/vectors.idx (binary Float32Array format)
Rebuild embeddings on compile

Search Improvements

Fuzzy matching for typo tolerance (edit distance ≤ 1 for tokens ≥ 4 chars)
Phrase search with quotes: kib search '"attention mechanism"'
Tag-based filtering: kib search "transformers" --tag deep-learning
Date range filtering: kib search --since 2024-01-01
Search result highlighting in terminal (bold matched terms)

v0.8.0 — Skill Ecosystem

Remote Skill Registry

kib skill install github:user/skill-name — install from GitHub repo
kib skill install <npm-package> — install from npm
Skill dependency resolution (skills can depend on other skills)
kib skill create <name> — scaffold a new skill from template
kib skill publish — publish to registry

Additional Built-in Skills

find-contradictions — detect contradictory claims across articles
weekly-digest — generate a weekly summary of new additions
export-slides — generate Marp slide deck from articles
timeline — generate chronological timeline from articles
compare — compare two articles/topics side by side
explain — explain a topic at a specified reading level
suggest-tags — auto-tag articles based on content analysis

Skill API Enhancements

Skill-to-skill invocation (one skill can call another)
Skill hooks: run automatically after compile, ingest, or lint
Skill configuration in config.toml
Skill output to specific wiki category

v1.0.0 — Production Ready

Reliability

Lockfile mechanism to prevent concurrent vault writes
Automatic backup before destructive operations (compile --force)
Crash recovery: detect incomplete writes and repair manifest
Validate manifest integrity on every load (detect corruption)

Documentation

docs/getting-started.md — quick start tutorial with real example
docs/vault-format.md — vault format specification
docs/skill-authoring.md — how to create custom skills
docs/provider-config.md — LLM provider setup guide
docs/architecture.md — codebase architecture for contributors
Example vaults in examples/ directory (ML research, software docs, reading list)
Blog post / launch announcement

Testing & Quality

E2E test suite: full init → ingest → compile → search → query with mock LLM
Performance benchmarks: measure compile time, search latency, cold start
CI: test on macOS, Linux, Windows
Code coverage reporting via bun test --coverage

Gamechangers

The features that take kib from "cool tool" to "can't live without it."

Passive Learning Daemon

kib should silently learn from everything you read without you thinking about it.

Chrome extension: "Send to KB" button + optional auto-capture of pages you spend >30s on
kib watch as a background daemon (launchd/systemd) — not just inbox, but browser history, clipboard, screenshots
OS-level integration: watch a folder of PDFs, auto-ingest Kindle highlights, Readwise sync
Zero-friction ingest: no commands, no thinking, it just absorbs

Instant Value Without Compile

Most of kib's value is locked behind kib compile. That's wrong — value should be immediate on ingest.

Search + query over raw sources directly (no compile required)
Compile becomes an optional enrichment step, not a prerequisite
Incremental indexing: search index updates on ingest, not compile
"Ask about this source" — query a single raw source without compiling the whole vault

Beyond CLI

CLI-only means developer-only. The knowledge is valuable to everyone.

VS Code extension: sidebar with search, query, ingest from editor
Obsidian plugin: sync kib vault ↔ Obsidian vault, use kib's compile + search
Web UI: local dashboard with graph visualization, search, query (not just export)
Raycast/Alfred integration: global hotkey → search your knowledge base
Mobile: read-only PWA for querying on the go

Shared Knowledge Bases

Personal wikis are useful. Team wikis are essential.

kib share — push vault to a git remote, team members clone + contribute
Multi-user ingest: team members ingest from their own browsers, shared compile
Access control: public wiki articles vs private notes
Team dashboard: who ingested what, what's new this week, knowledge gaps
Org-wide knowledge graph: connect team vaults into a federated search

Post v1.0 — Future Ideas

Vault Merging

kib merge <other-vault> — merge two vaults
LLM-powered conflict resolution for overlapping articles
Three-way merge: base + ours + theirs for article content
Merge strategies: "ours", "theirs", "combine", "llm-resolve"

Multi-Vault Support

kib vault list — list known vaults
kib vault switch <name> — switch active vault
Cross-vault search: kib search --vault all "query"
Cross-vault linking: [[vault:slug]] syntax

Real-Time Collaboration

Git-based async collaboration (vault is already git-friendly)
Conflict detection and resolution on git pull
Shared vault hosting (GitHub repo as vault backend)

Web UI

kib serve — local web server with read-only wiki viewer
Search interface in the browser
Article graph visualization (force-directed graph)
Reading mode with backlink sidebar

Additional Export Formats

PDF export (via Puppeteer or wkhtmltopdf)
Marp slide deck export
Anki flashcard deck export (.apkg)
Notion import/export
Obsidian vault compatibility layer

Telemetry (Opt-In)

Anonymous usage stats: which commands are used most, average vault size
Opt-in only, clearly documented
Self-hosted analytics endpoint

FilesExpand file tree

ROADMAP.md

Latest commit

History