The shared workspace for humans + AI

The first draft is AI. The final word is yours.

OMA is where AI agents make documents and decks—and people finish them. One file, open to both.

Connect your AI See how it works

Plain Markdown · MCP-native · Visual editing

Turn my Q3 notes into a six-slide reliability review

reliability-review.mdyou + agentsaved · v14

Q3 reliability

Reliability held as traffic grew

Availability cleared the target while request volume grew 2.4× year over year.

99.97%availability
SLO 99.9%

You

Now tighten the incident slide. Keep my headline.

Any MCP clientVersion-aware writesHuman-editable canvasPortable Markdown

One artifact. The whole conversation.

Ask. Edit. Ask again.

Platform reliability reviewAgent created 6 slidesv12 · saved

Q3 2026

Platform reliability review

What changed, what held, and what we owe next quarter.

99.97%availability · target 99.9%

Two ways in. One source of truth.

Markdown underneath. A real editor on top.

Agents get a format they understand. People get the canvas they expect. Neither side is working on a disposable copy.

reliability-review.mdsource

### Q3 reliability
# Reliability held as traffic grew

Availability cleared the target while
request volume grew 2.4× year over year.

```overlays
[
  { "type": "stat",
    "value": "99.97%",
    "slot": "right" }
]
```

Document canvaseditable

Q3 reliability

Reliability held as traffic grew

Availability cleared the target while request volume grew 2.4× year over year.

99.97% availability against a 99.9% SLO

Bring your own AI

No house model required.

If your agent speaks MCP over HTTP, it can work in OMA. The artifact stays the constant even when the model changes.

Claude CodeONE COMMAND TO CONNECT

Your internal agentSTREAMABLE HTTP

Any MCP clientSTANDARD TOOL SURFACE

The next modelSAME PORTABLE ARTIFACT

Power without the wreckage

Let the agent work. Keep the file safe.

01 / PRECISE

Small edits

Change a phrase, one slide, or a known position instead of rewriting the whole artifact.

02 / CURRENT

Version checks

Every write carries the version last read. Stale changes stop before they overwrite newer work.

03 / VALID

All-or-nothing saves

Bad overlays and unsupported diagrams reject the entire write with specific errors.

Developer quickstart

Connect once. Then ask for the work.

MCP gives an AI client a standard way to use OMA’s tools. The private pilot currently runs with the local app.

OMA / CONNECTMCP + SKILL + VERIFY

# Start OMA
make dev

# Connect Claude Code
claude mcp add --transport http --scope user \
  ohmyartifact http://localhost:8080/mcp

# Then ask
“Turn these notes into a six-slide decision deck.”

Set up ohmyartifact for me. Do all three steps, then tell me what happened.

1. Connect the MCP server. It is a Streamable HTTP server at http://localhost:8080/mcp (the app must be running: `make dev` inside ohmyartifact/app).
   - Claude Code: run `claude mcp add --transport http --scope user ohmyartifact http://localhost:8080/mcp`
   - Any other client: add {"ohmyartifact": {"type": "http", "url": "http://localhost:8080/mcp"}} under "mcpServers" in its MCP config.

2. Install a skill named ohmyartifact in the client's skills folder. Its instructions: ohmyartifact stores every artifact as plain markdown. A person edits it on a canvas; you edit it over MCP. Read the server-provided resource `oma://guide` before your first write. Use `list_artifacts` then `read_artifact`; keep the returned version. Pick the narrowest write tool. On a stale version, read again and retry once. Invalid content rejects the whole write. Report changes by heading or slide number.

3. Verify. Call `list_artifacts`, then read `oma://guide` and report its first heading. If either fails, explain why instead of retrying.

Read the full reference →

Private-pilot pricing

Pay for the work you keep. Not the people who use it.

Every plan includes unlimited users and human-plus-agent editing. Capacity is based on active saved artifacts.

Business

$299 / month

$2,990 yearly · two months included

1,000 artifacts active and saved

Scale

$799 / month

$7,990 yearly · two months included

10,000 artifacts active and saved

Enterprise

$1,999 / month

$19,990 yearly · two months included

50,000 artifacts active and saved

Checkout reserves the selected plan and starts assisted onboarding. Deleted artifacts do not count toward capacity.

Make the draft. Keep the authorship.

Give every AI a real place to work—and every person the final say.

Connect your AI Read the docs