Seam

Where ideas connect.
Where your notes become your agents' context.

---

You talk to a lot of agents. Claude Code for backend, Cursor for frontend, a research conversation, a debugging session. Each one is a separate context window that forgets everything when it closes. The reasoning, the dead ends, the decisions that didn't make it into a commit -- gone.

Memory layers try to fix this by giving agents a key-value store. Seam goes further: it's a full knowledge system -- notes, projects, tasks, wikilinks, a knowledge graph, daily briefings, autonomous organization -- that your agents work inside, not just read from. They search your notes, create new ones, manage projects, collaborate through shared research labs, and build on each other's findings. Your knowledge base and their context are the same plain .md files on disk.

Everything runs on your machine by default. Local AI, local storage, nothing leaves unless you want it to.

Seam -- the join between two pieces. Knowledge gains meaning at the intersections.

Why Seam

Not a memory layer -- a knowledge system. Memory layers store key-value pairs. Seam gives your agents a full workspace: notes with wikilinks and tags, projects, tasks, a knowledge graph, semantic search, daily briefings, and an autonomous librarian that organizes it all. Every MCP-compatible agent -- Claude Code, Cursor, Windsurf -- works inside the same system. What one agent learns, the next one knows.

Your notes are files. Not rows in someone else's database. Plain markdown with YAML frontmatter, organized in folders. Edit them with Seam, vim, VS Code, or anything else. Back them up with git, sync them with Syncthing, grep them from the terminal. No export step because there's nothing to export from.

AI runs on your machine. Ollama powers everything by default -- embeddings, chat, search, writing assist. Zero cloud dependencies, zero API costs, zero privacy trade-offs. Need more horsepower? Switch to OpenAI or Anthropic with one config line. Your notes stay local either way.

What You Can Do

Give your agents a workspace, not a scratchpad

Seam's MCP server exposes 40+ tools that let agents operate inside your knowledge base -- not just store memories, but search notes, create new ones, manage projects, track tasks, traverse the knowledge graph, and collaborate through shared research labs. Your afternoon debugging session in Cursor knows what your morning architecture session in Claude Code decided, because they worked in the same system.

Ask your notes anything

"What did I write about caching strategies?" works even if you never used the word "caching." Semantic search finds meaning, not just keywords. Ask Seam combines retrieval with generation to answer questions grounded in things you actually wrote, with citations back to specific notes. Or synthesize across up to 50 notes at once -- "Summarize everything I know about project X."

Work with an assistant that acts

The agentic assistant doesn't just answer questions about your knowledge base -- it works inside it. Ask it to capture a meeting summary, plan a project, find connections between ideas, or rewrite a section. It has 19 tools spanning notes, projects, tasks, search, the knowledge graph, and its own long-term memory. Six tools that write data pause and ask for your explicit approval. Every action is recorded in a full audit trail.

Capture fast, find later

Quick capture for passing thoughts. URL-to-note for articles. Voice transcription via local Whisper -- no audio leaves your machine. Daily notes. Templates. Everything lands in your inbox until you or the Librarian sorts it.

See how ideas connect

The knowledge graph shows connections between notes through [[wikilinks]], shared #tags, and projects. Related notes surface semantically similar content you forgot you wrote. Two-hop backlinks reveal indirect connections. Orphan detection finds notes that need linking.

Let AI organize for you

The Librarian is an autonomous background service that reviews orphaned and untagged notes, assigns them to projects, and adds tags from your existing taxonomy. It never touches your content, never invents new categories, and only processes notes that have been quiet for 15+ minutes. A library that shelves its own books.

Quick Start

git clone https://github.com/0spoon/seam.git
cd seam
make build          # builds bin/seamd (server), bin/seam (TUI), bin/seam-reindex (re-embed tool)
make init           # interactive setup: JWT secret, data dir, LLM provider, ChromaDB
make dev            # runs seamd + Vite + Chroma in parallel (Ctrl-C to stop)

Seam works without AI -- you get a solid markdown note system with full-text search out of the box. Add Ollama for AI features. Add ChromaDB for semantic search (make init can manage a Docker container for you). See Getting Started for prerequisites and the full setup walkthrough.

How It Works

You write .md files       Seam watches and indexes       AI connects the dots
       |                           |                            |
       v                           v                            v
    notes/                      seam.db                     ChromaDB
    inbox/                    (SQLite FTS5)                 (embeddings)
    project-a/               metadata, links               similarity
    project-b/              tasks, versions                  search
       |                           |                            |
       +------------+--------------+----------------------------+
                    |
              seamd (Go binary)
              |       |       |
            Web     TUI    MCP Server
          (React) (Bubble  (AI coding
                   Tea)    agents)

A single Go binary serves the REST API, WebSocket events, and the MCP endpoint. SQLite handles metadata and full-text search. Your .md files on disk are the source of truth -- Seam watches for external edits and re-indexes automatically. ChromaDB stores vector embeddings for semantic search (optional, degrades gracefully if absent).

Three interfaces to the same data: a web app, a terminal TUI, and a full REST API.

Graceful Degradation

Missing	What happens
Ollama	AI features disabled. Markdown notes + FTS5 search still work.
ChromaDB	No semantic search. Full-text search and AI chat still work.
Whisper	No voice capture. Everything else works.

Documentation

Document	Description
Getting Started	Prerequisites, installation, configuration
AI & Assistant	LLM providers, Ask Seam, agentic assistant, Librarian
Architecture	System diagram, tech stack, data format, project layout
API Reference	REST endpoints, WebSocket events
MCP Server	Persistent memory and knowledge base access for AI coding agents
Development	Build, test, lint, project structure
Security	Threat model, auth, input validation, assistant safety
Brand Guidelines	Visual identity, colors, fonts, logo usage

License

MIT