CLI tool for managing knowledge and documentation in Git + Markdown. Sync, search (full-text + vector), AI assistance, task tracking, and traceability between documentation and code — all in one binary, no servers or databases.

# Debug
just build

# Release
just release

cargo run -- <command> [args]
./target/release/kqs <command>

just test
cargo test --workspace -- --nocapture

kqs runs in two contexts:

# Explicit
kqs --dev check scan
kqs --doc doc new adr "Title"

# Auto-detect: if docs/ + TypeSpec/ exist → doc mode
kqs check traceability

# Via env
KQS_MODE=dev kqs check scan

In dev mode, doc-only commands (kqs doc new, kqs typespec new, kqs readme) fail with an explanatory error.

# Create a knowledge repository
kqs init
kqs init --path ~/my-knowledge --remote https://github.com/user/knowledge.git

# List templates
kqs doc template --list

# Create a document
kqs doc new adr "Database Choice"      # → docs/03-architecture/adr-001-*.md
kqs doc new tz "API Gateway Design"    # → docs/04-technical-design/tz-001-*.md
kqs doc new bft "Customer Portal"      # → docs/01-business-foundation/bft-001-*.md
kqs doc new frd "Sign-up"              # → docs/02-product-ux/frd-001-*.md

# Screen and User Flow
kqs screen "Login Screen"
kqs userflow "Onboarding Flow"

# List all documents
kqs doc list

Every template in .kqs/templates/ carries a specific semantic load — the type is chosen by what question the document answers, not by preference. The types form a chain: each next type refines the previous one, from business intent down to the data specification that code implements.

Create documents in this order — each type flows from the previous one:

idea → bft → brd/frd/nfr → adr → rfc → tz → typespec → code

1. idea — a thought worth exploring. Once validated, formalize it into a bft.
2. bft — the root of the chain: what the business wants and why (objectives, stakeholders, scope). Example: "User Authentication".
3. brd / frd / nfr — derived from the bft: business, functional and non-functional requirements that detail it.
4. adr — an architectural decision made to satisfy those requirements: context, decision, consequences.
5. rfc — a proposal for a significant change, discussed before it becomes a final design.
6. tz — the technical design that fixes the decision in concrete terms: architecture, API, data model.
7. typespec — the data specification: models the design as checkable TypeSpec contracts.
8. code — implements the specification, linked back via @doc-anchor.

kqs check traceability-deep verifies the chain is complete (BFT → ADR → TZ → TypeSpec), flagging any missing link.

# Create a model
kqs typespec new User
# → TypeSpec/User.tsp, updated TypeSpec/main.tsp

# List models
kqs typespec list
# → Name    File        Doc Refs
# → User    User.tsp    TZ-001

# Full report
kqs check traceability
# → Table: Object, Type, Status, Links, Recommendation

# Orphans (TypeSpec without documentation)
kqs check orphans

# Full chain BFT → FRD → ADR → TZ → TypeSpec
kqs check traceability-deep --deep

# Custom chain
kqs check traceability-deep --deep --chain "bft adr tz"

# JSON for editor plugins
kqs check traceability-deep --deep --json

Deep coverage matrix:

• ✅ covered — full chain ❌ broken — chain gap
• 👻 orphan — no incoming links 🕸️ stale — doc changed, code not updated

# Scan projects from the config for @doc-anchor
kqs check scan

# Rebuild the graph before scanning
kqs check scan --rebuild

When an orphan anchor is found (@doc-anchor PushNotificationManager without TypeSpec), a TASK with instructions is created automatically.

# All stale links
kqs check notify

# Last 7 days
kqs check notify --since 7d

knowledge_path = "."

[[projects]]
path = "../mobile-app"
label = "iOS App"

[[projects]]
path = "../backend"
label = "Backend"

[watcher]
debounce_secs = 600
ignore_patterns = []

[search]
chunk_size = 512
chunk_overlap = 64
max_results = 10
relevance_threshold = 0.7

# Regenerate
kqs readme

Generates between the <!-- kqs:start --> / <!-- kqs:end --> markers:

1. Kanban Board — tasks by status (todo/in_progress/review/done)
2. Table of Contents — document tree with clickable links

Updated automatically on kqs push and kqs watch.

kqs task new --title "Implement search" --priority high
kqs task list
kqs task status TASK-001 --set in_progress
kqs task show TASK-001
kqs task search "search"

# Full-text (FTS5, always available)
kqs search "system architecture"

# Vector (sqlite-vec + Candle, when a model is available)
kqs search --vector "semantic search"

# Hybrid (default)
kqs search "database" --limit 20

# Auto-commit changes
kqs watch

# Custom debounce
kqs watch --debounce-secs 120

# Trace daemon: watcher + auto-trace on every cycle
kqs watch --trace

kqs push                          # pull --rebase → readme-gen → push
kqs push --no-readme              # skip README generation
kqs push --dry-run                # show changes without pushing

Developers annotate code in their projects:

// @doc-anchor SecureTokenStorage
// @see docs://architecture/ADR-001.md
class KeychainStorage: TokenStorable { }

// @doc-anchor AuthServiceImpl
func ValidateToken(token string) bool { }

When scanning, kqs finds these annotations and builds the graph: BFT-001 → ADR-001 → TZ-001 → TypeSpec → @doc-anchor → code

Orphan anchors (anchor exists, no TypeSpec) → task created automatically.

kqs conflict list
kqs conflict show docs/architecture.md
kqs conflict resolve docs/architecture.md --ours
kqs conflict resolve docs/architecture.md --theirs

kqs ask "How is the architecture organized?"
kqs ask "Describe the User model" --insert docs/user-model.md

Requires an LLM provider in knowledge.toml (Ollama/OpenAI/Anthropic).

just lint        # clippy
just fmt         # fmt --check
just fmt-fix     # fmt
just check       # build + lint + test + fmt

# On every push/PR — build + clippy + test for macOS/Linux/Windows
.github/workflows/ci.yml

# On tag v* — release build for 3 platforms + GitHub Release
.github/workflows/release.yml

# On tag v* — publish all workspace crates to crates.io
.github/workflows/publish.yml

# From crates.io (installs the `kqs` binary)
cargo install kqs

# From source
cargo install --path kqs

kqs --help

Note on the name: the crates.io package kq is an unrelated KDL query tool. Our package is kqs and installs the kqs binary — no conflict with anything on crates.io.

Or download a binary from GitHub Releases.

Workspace crates are published to crates.io from a v* tag:

1. just bump-version 0.2.0 — bump the single workspace version in Cargo.toml
2. Commit and push the tag v0.2.0 (must match the workspace version)
3. CI runs tests, then publishes kq-config → kq-llm → kq-embeddings → kq-core → kqs in dependency order
4. just publish-check — local pre-flight: dry-run publish of the leaf crates and a package file audit of the dependents

Requires the CARGO_REGISTRY_TOKEN secret in GitHub repository settings.

kqs/               # CLI entry point (clap)
kq-core/           # Business logic
├── check.rs       # Traceability, orphans, deep coverage
├── code_anchor.rs # @doc-anchor scanner
├── db.rs          # SQLite: files, trace_nodes, trace_links, code_anchors
├── docs.rs        # DocType, DocNode, front-matter parser
├── git.rs         # Git operations
├── readme_gen.rs  # README generator (Kanban + TOC)
├── task.rs        # CRDT tasks
├── typespec.rs    # TypeSpec models
└── watcher.rs     # File watcher
kq-embeddings/     # Candle + all-MiniLM-L6-v2
kq-llm/            # Ollama, OpenAI, Anthropic providers
kq-config/         # knowledge.toml parsing

<project>/
├── .kqs/
│   ├── knowledge.toml      # config
│   ├── knowledge.db        # SQLite: FTS + vectors + trace graph
│   ├── events/             # CRDT task events
│   └── templates/          # document templates
├── users/                  # users (username.md)
├── docs/                   # documentation (8 categories)
├── tasks/                  # tasks (TASK-NNN.md)
├── TypeSpec/               # TypeSpec models
└── README.md               # auto-Kanban + Table of Contents

Tasks use CRDT events instead of overwrites:

create       → .kqs/events/TASK-001/20260709T100000Z-create.md
assign alex  → .kqs/events/TASK-001/20260709T110000Z-assign-alex.md
move review  → .kqs/events/TASK-001/20260709T120000Z-move-review.md

Why: no merge conflicts, status is computed by replay, readable git diff, parallel work without collisions.

Alice: kqs task assign TASK-001 alex  → ...assign-alex.md
Bob:   kqs task assign TASK-001 maria → ...assign-maria.md

After a rebase, both event files end up side by side; last-writer-wins is resolved by timestamp — no merge conflict.