engraph turns your markdown vault into a searchable knowledge graph that any AI agent can query — Claude Code via MCP, ChatGPT via Actions, or any tool via REST API. It combines semantic embeddings, full-text search, wikilink graph traversal, temporal awareness, and LLM-powered reranking into a single local binary. Same model stack as qmd. No API keys, no cloud — everything runs on your machine.
Local knowledge graph for AI agents. Hybrid search + MCP server for Obsidian vaults.
ดาว
172
เครื่องมือ
—
อัปไทม์
—
Why engraph?
Plain vector search treats your notes as isolated documents. But knowledge isn't flat — your notes link to each other, share tags, reference the same people and projects. engraph understands these connections.
5-lane hybrid search — semantic embeddings + BM25 full-text + graph expansion + cross-encoder reranking + temporal scoring, fused via Reciprocal Rank Fusion. An LLM orchestrator classifies queries and adapts lane weights per intent. Time-aware queries like "what happened last week" or "March 2026 notes" activate the temporal lane automatically.
MCP server for AI agents — engraph serve exposes 25 tools (search, read, section-level editing, frontmatter mutations, vault health, context bundles, note creation, PARA migration, identity) that Claude, Cursor, or any MCP client can call directly.
HTTP REST API — engraph serve --http adds an axum-based HTTP server alongside MCP with 26 REST endpoints, API key authentication, rate limiting, and CORS. Web-based agents and scripts can query your vault with simple curl calls.
Section-level editing — AI agents can read, replace, prepend, or append to specific sections by heading. Full note rewriting with frontmatter preservation. Granular frontmatter mutations (set/remove fields, add/remove tags and aliases).
Vault health diagnostics — detect orphan notes, broken wikilinks, stale content, and tag hygiene issues. Available as MCP tool and CLI command.
Real-time sync — file watcher keeps the index fresh as you edit in Obsidian. No manual re-indexing needed.
Smart write pipeline — AI agents can create, edit, rewrite, and delete notes with automatic tag resolution, wikilink discovery, and folder placement based on semantic similarity.
Fully local — llama.cpp inference with GGUF models (~300MB mandatory, ~1.3GB optional for intelligence). Metal GPU-accelerated on macOS (88 files indexed in 70s). No API keys, no cloud.
What problem it solves
You have hundreds of markdown notes. You want your AI coding assistant to understand what you've written — not just search keywords, but follow the connections between notes, understand context, and write new notes that fit your vault's structure.
Existing options are either cloud-dependent (Notion AI, Mem), limited to keyword search (Obsidian's built-in), or require you to copy-paste context manually. engraph gives AI agents direct, structured access to your entire vault through a standard protocol.
Index — walks your vault, chunks markdown by headings, embeds with a local GGUF model via llama.cpp (Metal GPU on macOS), stores everything in SQLite with FTS5 + sqlite-vec + a wikilink graph
Search — an orchestrator classifies the query and sets lane weights, then runs up to five lanes (semantic KNN, BM25 keyword, graph expansion, cross-encoder reranking, temporal scoring), fused via RRF
Serve — starts an MCP server that AI agents connect to, with a file watcher that re-indexes changes in real time
Quick start
Install:
# Homebrew (macOS)
brew install devwhodevs/tap/engraph
# Pre-built binaries (macOS arm64, Linux x86_64)
# → https://github.com/devwhodevs/engraph/releases
# From source (requires CMake for llama.cpp)
cargo install --git https://github.com/devwhodevs/engraph
Index your vault:
engraph index ~/path/to/vault
# Downloads embedding model on first run (~300MB)
# Incremental — only re-embeds changed files on subsequent runs
Search:
engraph search "how does the auth system work"
1. [97%] 02-Areas/Development/Auth-Architecture.md > # Auth Architecture #6e1b70
OAuth 2.0 with PKCE for all client types. Session tokens stored in HTTP-only cookies...
2. [95%] 01-Projects/API-Design.md > # API Design #e3e350
All endpoints require Bearer token authentication. Tokens are issued by the OAuth 2.0...
3. [91%] 03-Resources/People/Sarah-Chen.md > # Sarah Chen #4adb39
Senior Backend Engineer. Tech lead for authentication and security systems...
Note how result #3 was found via graph expansion — Sarah's note doesn't mention "auth system" directly, but she's linked from the auth architecture doc via [[Sarah Chen]].
Claude Code — Install the plugin (recommended):
claude plugin marketplace add devwhodevs/engraph
claude plugin install engraph@engraph
Connect to Claude Code:
Or configure MCP manually in ~/.claude/settings.json:
Now Claude can search your vault, read notes, build context bundles, and create new notes — all through structured tool calls.
AI Agent Skills — Install the Engraph skills using the skills CLI (recommended):
npx skills add devwhodevs/engraph
Enable HTTP REST API:
# Start MCP + HTTP server on port 3000
engraph serve --http
# Custom port and host
engraph serve --http --port 8080 --host 0.0.0.0
# Local development without API keys (127.0.0.1 only)
engraph serve --http --no-auth
API key management:
# Add a new API key (read or write permission)
engraph configure --add-api-key
# List existing keys
engraph configure --list-api-keys
# Revoke a key
engraph configure --revoke-api-key eg_abc123...
engraph search "how does authentication work" --explain
1. [97%] 01-Projects/API-Design.md > # API Design #e3e350
All endpoints require Bearer token authentication...
Intent: Conceptual
--- Explain ---
01-Projects/API-Design.md
RRF: 0.0387
semantic: rank #2, raw 0.38, +0.0194
rerank: rank #2, raw 0.01, +0.0194
02-Areas/Development/Auth-Architecture.md
RRF: 0.0384
semantic: rank #1, raw 0.51, +0.0197
rerank: rank #4, raw 0.00, +0.0187
The orchestrator classified the query as Conceptual (boosting semantic lane weight). The reranker scored each result for relevance as the 4th RRF lane.
engraph resolves tags against the registry (fuzzy matching), discovers potential wikilinks ([[Sarah Chen]]), suggests the best folder based on semantic similarity to existing notes, and writes atomically.
Edit a specific section:
engraph write edit --file "Meeting Notes" --heading "Action Items" --mode append --content "- [ ] Follow up with Sarah"
Targets the "Action Items" section by heading, appends content without touching the rest of the note.
Returns orphan notes (no links in or out), broken wikilinks, stale notes, and tag hygiene issues.
HTTP REST API
engraph serve --http adds a full REST API alongside the MCP server, exposing the same capabilities over HTTP for web agents, scripts, and integrations.
Keys have either read or write permission. Write keys can access all endpoints; read keys are restricted to read-only endpoints. Use --no-auth for local development without keys (127.0.0.1 only).
engraph migrate para restructures your vault into the PARA method (Projects, Areas, Resources, Archive) using heuristic classification. The workflow is non-destructive: preview first, review the plan, then apply.
Workflow:
# 1. Preview — classify notes and generate a migration plan
engraph migrate para --preview
# Outputs: markdown summary + JSON plan saved to ~/.engraph/
# 2. Review the plan (edit if needed)
cat ~/.engraph/migration_preview.md
# 3. Apply — move files according to the plan
engraph migrate para --apply
# 4. Undo — reverse the last migration if something looks wrong
engraph migrate para --undo
Classification signals:
Category
Detection signals
Projects
Open tasks (- [ ]), active/in-progress status in frontmatter, project tags
You are a knowledge assistant connected to the user's Obsidian vault via engraph.
WORKFLOW:
1. Use searchVault to find relevant notes before answering questions
2. Use readNote for full content, readSection for specific headings
3. Use getWho for people context, getProject for project context
4. Use getVaultMap to orient yourself in the vault structure
5. Only create or edit notes when explicitly asked
SEARCH TIPS:
- Temporal queries ("last week", "yesterday") activate time-aware search automatically
- Results include confidence % — prefer higher confidence matches
- Fuzzy matching works: typos in names are handled
STYLE:
- Reference vault notes by name when answering
- Quote relevant snippets
- If information isn't in the vault, say so clearly
- Be concise
Click Add Action → Import from URL
Enter: https://your-tunnel-url/openapi.json
Click the gear icon next to Authentication
Select API Key, Auth Type: Bearer
Paste your API key (the eg_... key from Step 1)
Save and test
Conversation starters
"What happened in my vault last week?"
"Summarize my current work projects"
"Find notes related to [topic]"
"Create a note about today's meeting with [person]"
Notes
Tunnel URLs are temporary (Cloudflare quick tunnels change on restart). For persistent URLs, set up a named Cloudflare tunnel or use ngrok with a reserved domain.
Read-only mode: set permissions = "read" on the API key if you don't want ChatGPT to create or modify notes.
Rate limiting: default is 60 requests/minute per key. Adjust rate_limit in config if needed.
engraph must be running on your machine for ChatGPT to access it. If you close the terminal, the connection drops.
Use cases
AI-assisted knowledge work — Give Claude or Cursor deep access to your personal knowledge base. Instead of copy-pasting context, the agent searches, reads, and cross-references your notes directly.
Developer second brain — Index architecture docs, decision records, meeting notes, and code snippets. Search by concept across all of them.
Research and writing — Find connections between notes that you didn't explicitly link. The graph lane surfaces related content through shared wikilinks and mentions.
Team knowledge graphs — Index a shared docs vault. AI agents can answer "who knows about X?" and "what decisions were made about Y?" by traversing the note graph.
Temporal search: natural language date queries ("last week", "March 2026", "recent"), date extraction from frontmatter and filenames, smooth decay scoring
Confidence % display: search results show normalized 0-100% confidence instead of raw RRF scores
LLM research orchestrator: query intent classification + query expansion + adaptive lane weights
llama.cpp inference via Rust bindings (GGUF models, Metal GPU on macOS, CUDA on Linux)
Intelligence opt-in: heuristic fallback when disabled, LLM-powered when enabled
MCP server with 25 tools (8 read, 10 write, 2 identity, 1 index, 1 diagnostic, 3 migrate) via stdio
HTTP REST API with 26 endpoints, API key auth (eg_ prefix), rate limiting, CORS — enabled via engraph serve --http
User identity with L0/L1 tiered context for AI agent session starts
Section-level reading and editing: target specific headings with replace/prepend/append modes
Full note rewriting with automatic frontmatter preservation
Granular frontmatter mutations: set/remove fields, add/remove tags and aliases
Soft delete (archive) and hard delete (permanent) with audit logging
Vault health diagnostics: orphan notes, broken wikilinks, stale content, tag hygiene
Obsidian CLI integration with circuit breaker (Closed/Degraded/Open) for resilient delegation
Real-time file watching with 2s debounce, startup reconciliation, and watcher coordination to prevent double re-indexing
Write pipeline: tag resolution, fuzzy link discovery, semantic folder placement
Context engine: topic bundles, person bundles, project bundles with token budgets
Vault graph: bidirectional wikilink + mention edges with multi-hop expansion
Placement correction learning from user file moves
Enhanced file resolution with fuzzy Levenshtein matching fallback
Content-based folder role detection (people, daily, archive) by content patterns
PARA migration: AI-assisted vault restructuring into Projects/Areas/Resources/Archive with preview, apply, and undo workflow
Configurable model overrides for multilingual support
426 unit tests, CI on macOS + Ubuntu
Roadmap
Research orchestrator — query classification and adaptive lane weighting (v1.0)
LLM reranker — optional local model for result quality (v1.0)
MCP edit/rewrite tools — full note editing for AI agents (v1.1)
Vault health monitor — orphan notes, broken links, stale content, tag hygiene (v1.1)
Obsidian CLI integration — auto-detect and delegate with circuit breaker (v1.1)
Temporal search — find notes by time period, date-aware queries (v1.2)
HTTP/REST API — complement MCP with a standard web API (v1.3)
PARA migration — AI-assisted vault restructuring with preview/apply/undo (v1.4)
