termynl documentation

The terminal-native work platform. Email · docs · tasks · chat · agents on one object model.

What is termynl

termynl puts your whole workday behind a single command bar. Tasks, notes/docs, email, chat, and AI agents all run through one object model and one change bus, so the connective tissue — backlinks, activity, notifications, real-time, and AI actions — is written once and works everywhere. It's terminal-native, keyboard-first, and self-hostable.

Getting started

Sign in. Microsoft (work/school), Google, GitHub, or an email magic-link. New sign-ins start a 14-day trial.
Meet the bar. Everything happens from the bottom command bar — create, link, search, run an agent.
Create a task: press :, type task Ship the billing epic, Enter.
Write a note: open docs, press + new note, and just type. Press / for blocks and @ to link.

Keyboard & commands

termynl is 100% keyboard-driven with a non-Ctrl leader.

Key	Does
`:`	Inline command line — the leader. Type a command and run it.
`/`	Command palette (and, in the editor, the block menu).
`@`	Mention/link a task, note, or person — inline, by keystroke.
`1`–`7`	Jump straight to a surface (agents, email, docs, tasks, …).
`Tab` / `⇧Tab`	Cycle surfaces.
`?`	Show the shortcuts overlay. `esc` closes anything.

One object model

Every entity — a task, note, contact, message — is a typed object row. Nothing mutates directly: every change is a tx appended to a log, and a document's current state is a fold over that log. That append-only log is the source of truth, the audit trail, and the real-time event stream at once.

Because every write flows through the same pipeline (validate → authorize → persist → triggers → broadcast), features that would normally be N×N integrations are written once: mention a task from a note and the backlink just exists; an agent moving a task produces the same activity a human would.

Links & the graph

@-mentions are structured nodes, never regex over text. Each creates a bidirectional reference edge, so every task and note shows what links to it and what it links to. The connections surface renders the live graph of those edges.

Notes → disk (live vault)

Every note body live-syncs to a markdown file on disk, at a scope-encoded path (workspace/space/<note>.md) with YAML frontmatter. That means Claude Code and your agents read a real, current vault of your workspace — the same LLM-Wiki pattern, but maintained automatically.

Import from Obsidian or Notion

Bring your existing knowledge base — links become live backlinks on the way in.

Obsidian

Point the importer at your vault. [[wikilinks]] and [[link|alias]] resolve to mention nodes (two-pass, so forward links work; links outside the import stay plain text). YAML frontmatter → properties, #tags → tags.

Notion

Export your workspace as Markdown & CSV and import it. termynl strips Notion's page-id suffixes for clean titles and resolves [label](Page id.md) internal links to backlinks; external URLs stay as links.

Agents & MCP

termynl exposes an MCP endpoint at api.termynl.com/mcp. Any agent — Claude Code, or your own Hermes/Mira — drives termynl through tools (create_task, create_note, find_items, …).

Crucially, tools author changes as an agent account through the same pipeline as a human — inheriting space permissions and producing activity/backlinks. There is no privileged side-door, so an agent can never see or touch a space its account can't access.

POST /mcp   { "jsonrpc":"2.0", "id":1, "method":"tools/call",
              "params":{ "name":"create_task",
                         "arguments":{ "space":"space-product", "title":"…" } } }

Bring your own model (Anthropic by default; OpenAI/local) or connect a named Hermes agent as a termynl channel.

Storage — bring your own

On the hosted service you choose your file storage at setup: OneDrive, Google Drive, or a 3rd-party S3 bucket. termynl stores only the index (metadata, references, search text, embeddings) — your file bytes live in your own cloud. Email and calendar are viewed live from Microsoft/Google, never bulk-copied.

Billing

$35 per seat / month, billed on active members, after a 14-day trial. Manage payment and cancel in the self-service portal. No free hosted tier; the single-tenant build is free and open-source (see self-hosting).

Self-hosting

The single-tenant app is open-source and runs on your own infrastructure — your data never leaves your box. Deploy with Docker (an updater sidecar keeps it current). Multi-tenant orchestration and billing are the proprietary control-plane and aren't part of the self-host build.

More self-hosting detail (compose files, updater, backups) lands here as the open-core build is finalized.

Questions? [email protected] · termynl.com