OneMira C1 Docs

OneMira C1 is a local context device that records conversations, transcribes them on-device, and exposes the data to your AI coding tools via MCP. Run mira mcp configure to connect your editor in one command. This page covers setup, MCP configuration, tools reference, prompt recipes, CLI commands, and troubleshooting.

What is OneMira C1?

Features

Capture by side button, Auto-Capture on Launch, or scheduled work-hour windows.
Two MEMS microphones with on-chip AI noise canceling and about 3 m (10 ft) close-range pickup.
Records the far side of remote calls without taking over your mic — even with headphones on.
Works fully offline; useful speech becomes local Context and audio is deleted after sync.

After setup, supported AI tools on your desktop can access this context via the MCP protocol. Use Mira through MCP tools, slash commands (such as /mira-search), or tool references (such as @mira_search) depending on your AI client.

💡

Natural Language Discovery — This follows the Agent Skills open standard (adopted by Claude, Codex, and Cursor). When an agent starts, it only reads each capability's name, a one-sentence description, and trigger conditions. You don't need to memorize exact command formats like /mira-search or @mira_search; simply ask in plain natural language, and capable agents will automatically discover and invoke the tools.

🔒

Privacy by design — No required OneMira cloud workspace, cloud transcription service, or subscription for the core workflow.. Local transcription runs on your computer. Set up once, use forever.

What's in the Box?

OneMira C1 × 1

USB-C Cable × 1

Button & Port

OneMira C1 has one main control, two microphones, a USB-C port, and visual feedback elements.

💡

The turntable turns slowly as it listens — a calm, visible cue to everyone in the room that capture is on. The motion isn't decoration; it's how C1 keeps recording honest.

How Do I Set Up My C1?

The desktop app walks you through every step. The entire process takes about 5 minutes.

Install the OneMira Desktop App

Download and install the OneMira desktop application on your Mac or Windows computer:

Go to Downloads →

Activate Your Device

Launch the OneMira desktop app.
Press and hold the Multi-Function Button for 3 seconds until the LED indicator flashes white, indicating that Bluetooth pairing mode is active.
Follow the instructions in the app to activate your device.

321

Press and hold for 3 seconds — the app and device show pairing status

💡

Connect multiple machines to the same C1 device over MCP to share a single local context store..

Deploy the Transcription Engine

Local ASR model runs on the paired desktop app; no internet is needed after the initial download..

Choose your transcription language: Auto-Detect, Chinese, or English..
Click "Deploy Transcription Engine". The download is typically 100-500 mb depending on the model..
Wait for the progress bar to reach 100%. The app shows real-time download logs and a desktop illustration with the deployment percentage.

Connect AI Tools

MCP (Model Context Protocol) is how your AI tools can call local context from the physical world. Claude, Codex, Cursor, Claude Desktop, and MCP-compatible tools can use the same local files, CLI workflow, and MCP interface..

The app automatically scans your computer for supported AI tools.
Follow the app's AI access flow to write the configuration into each detected tool.

Claude Desktop

Claude Code

Codex

Cursor

other MCP-compatible tools.

⚡

Manual setup? If your AI tool isn't auto-detected, go to Settings → AI Access to copy the MCP JSON snippet and paste it into your tool's MCP configuration manually.

How Do I Use My C1 Daily?

Slow, visible turntable rotation indicates active recording.. Scroll through each mode below to see how the device behaves.

Recording

Press the side button once to start recording — the white LED lights up and the turntable spins. Press again to stop and return to standby. Click the button below to see it in action.

Standby: LED off, device is idle.

Low Battery

When battery drops below 10 % during recording, the LED switches from white to red. Charge before the next session.

Red LED indicates low battery while recording.

Charging

Plug in USB-C while idle — a red LED stays on during charging. When fully charged, the LED turns white.

Red LED while charging; white LED when full.

Understanding MCP

Before configuring your tools, it helps to understand what MCP is, why it matters for a local-first device like OneMira C1, and how the pieces fit together.

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets AI assistants call local tools and read local data — without sending anything to the cloud. Think of it as a USB port for AI: any MCP-compatible agent (Claude Desktop, Claude Code, Codex, Cursor, and 20+ others) can plug in and query your OneMira C1 data instantly.

💡

No API keys required — MCP servers run as local processes on your machine. Your AI assistant talks to them over stdin/stdout, just like a shell command. Nothing leaves your computer.

Why Local-First MCP?

Cloud RAG (Retrieval-Augmented Generation) requires you to upload transcripts, wait for indexing, and pay per query. OneMira C1 inverts this model:

Private — Transcripts stay on your machine. Audio is deleted after sync.
Instant — No upload latency. SQLite full-text search returns results in milliseconds.
Portable — Your data lives in plain Markdown files you can version, back up, or move.
Free forever — No subscription. No per-seat fees. No cloud dependency.

Architecture Overview

Here is how the data flows from your real-world conversations to your AI tools:

architecture

# Data Flow: Conversation → Context → AI Action

OneMira C1 Hardware → BLE Sync → Desktop App

↓

Local ASR Engine

(Whisper / mlx-whisper)

↓

Transcripts → SQLite DB

Wiki Vault → Markdown

↓

MCP Server (mira serve)

stdin/stdout transport

↓

AI Agent (Cursor / Claude Code / VS Code / ...)

# Your agent calls mira_search, mira_sessions, etc. to query local Context

MCP Setup

OneMira exposes local conversation context data to your AI tools via the Model Context Protocol (MCP). Below you will find the CLI command reference to automatically manage your integrations, and details to manually configure various editors.

Automatic Setup (CLI Tools)

The OneMira desktop application includes the mira Command Line Interface (CLI) to easily manage MCP client settings on your computer.

mira-cli

$ mira mcp list

# Scan and list all installed AI clients and their status

$ mira mcp configure <client_id>

# Automatically inject OneMira MCP server config into the client (omit client_id for interactive menu)

$ mira mcp unconfigure <client_id>

# Remove OneMira configuration from the client

Manual Configuration Guide

If your AI client is not auto-detected, or if you prefer to edit settings yourself, choose your tool from the complete list below to view the default paths and configuration.

Choose your AI platform:

macOS Config Path:~/Library/Application Support/Claude/claude_desktop_config.json

Windows Config Path:%APPDATA%/Claude/claude_desktop_config.json

Add the following block to your configuration file:

{
  "mcpServers": {
    "mira": {
      "command": "mira",
      "args": ["serve", "--mcp"]
    }
  }
}

Official Docs →Note: Restart Claude Desktop for changes to take effect.

Remote Access via SSE

Need to access your OneMira C1 meeting data from a cloud-based AI tool like Claude.ai or other remote agent workflows? The Mira MCP server supports a remote mode using Server-Sent Events (SSE) transport, which tunnels your local context through a secure connection.

remote-access

# Start the MCP server in SSE mode for remote AI clients

$ mira mcp serve --mcp

# The server exposes an SSE endpoint that cloud AI tools can connect to

# Your transcripts stay local — only query responses are sent over the wire

🔒

Data stays local — Even in remote mode, the MCP server runs on your machine. Cloud AI tools send queries over the tunnel; only the requested results (not raw transcripts or audio) are returned.

MCP Tools Reference

Once connected, your AI assistant automatically discovers and can call the following tools. They are grouped by function for easy reference.

Search & Discovery

These tools let your AI agent find information across transcripts, wiki pages, and meeting metadata.

Tool	Description
`mira_search`	Full-text search across both meeting transcripts and wiki notes. Use when the user asks what was discussed, decided, or mentioned.
`mira_guide`	Describe your intent in natural language; returns recommended tool calls with parameters. A router for agents unfamiliar with the tool set.
`mira_stats`	Get meeting statistics — session count, total duration, annotations, and open tasks at a glance.
`mira_sync_status`	Snapshot of the BLE device sync pipeline — streaming, transcribing, and pending-upload counts.

Session Management

Manage meeting sessions — browse, create, edit transcripts, and organize your recording library.

Tool	Description
`mira_sessions`	Browse sessions: list recent sessions (filter by date range, source, state, pin/star), or show a single session with its transcript preview.
`mira_segments`	Read exact transcript segments from a session, with optional timestamp range or tail limit for precision retrieval.
`mira_session_create`	Create a note session from text or file content. Orchestrates create → retitle → append → finalize in one call.
`mira_session_edit`	Overwrite a session's entire finalized transcript with new markdown content.
`mira_session_retitle`	Rename a session's title (non-destructive, text-only change).
`mira_session_pin`	Toggle a session's pinned state to keep important meetings easily accessible.
`mira_session_star`	Toggle a session's starred state for quick favorites access.
`mira_session_delete`	Delete a session (requires `confirm=true`; omit for a safe preview first).
`mira_session_prune`	Delete sessions older than N days (keeps pinned/starred). Requires confirmation.
`mira_segment_speaker`	Assign a speaker name to transcript segments within a time range.

Wiki & Knowledge

Read, write, and navigate your local Markdown knowledge vault — the persistent, structured output of your meetings.

Tool	Description
`mira_wiki_list`	List all wiki pages across all kinds (session, topic, person, project) with summaries.
`mira_wiki_read`	Read a wiki page by path (e.g. `topics/pcb-design`, `people/alice`). Returns kind, title, and content.
`mira_wiki_write`	Create or overwrite a wiki page. Content must include YAML frontmatter.
`mira_wiki_delete`	Delete a wiki page by path and update all related indexes.
`mira_wiki_backlinks`	List pages that reference a given wiki page via `[[wikilinks]]`.
`mira_wiki_sessions`	List session IDs whose transcripts reference a wiki page.
`mira_restore_version`	Restore a wiki page or session markdown to a historical version from git history.

Admin & Maintenance

System-level tools for backups, MCP client management, and AI job orchestration.

Tool	Description
`mira_mcp_list`	List MCP-capable AI editors with detection and configuration status.
`mira_mcp_configure`	Inject Mira's MCP config into a target AI editor automatically.
`mira_mcp_unconfigure`	Remove Mira's MCP config from a target editor (leaves other configs intact).
`mira_mcp_snippet`	Return a JSON snippet for manual paste into any editor config file.
`mira_admin_export_backup`	Export your entire library as a backup zip file.
`mira_admin_import_dry_run`	Preview what applying a backup zip would change (merge or replace mode).
`mira_admin_import_apply`	Apply a backup zip to your library (use dry run first).
`mira_ai_job_list`	List pending AI background jobs (ingest, synthesize, lint).
`mira_ai_job_drain`	Drain one pending AI job from the queue. Enables external agents to trigger ingest.

Skills & Prompt Recipes

OneMira ships with Agent Skills — pre-built instruction packages that teach your AI assistant how to use Mira's tools effectively. Skills follow the open Agent Skills standard adopted by Claude, Codex, Cursor, and others.

Built-in Skills

When you run mira mcp configure, Mira automatically installs three skills into your AI client. Each skill is a Markdown file that your agent reads on startup — no code, just instructions:

Skill	Trigger	What It Does
`mira-search`	"What was discussed about…"	Guides the agent to call `mira_search` with the right query structure, then format results with session links and timestamps.
`mira-brief`	"Prepare a brief for…"	Collects recent sessions, decisions, and open tasks about a topic or person into a concise briefing document.
`mira-todos`	"What are my action items?"	Reads `vault/tasks.md` and filters open items by owner, due date, or project.

💡

You don't need to memorize commands — Skills teach your AI to discover the right tools automatically. Just ask in plain English.

Prompt Recipes

Try these proven prompts with any MCP-connected AI client. Copy-paste them directly, or adapt them to your workflow:

prompt-recipes

# Meeting Recall

"What was decided about the database migration in last week's meetings?"

# Standup Prep

"Prepare a brief for tomorrow's standup. Include what I discussed yesterday,

open action items assigned to me, and any blockers mentioned."

# Stakeholder Context

"Summarize everything Sarah mentioned about the Q3 roadmap across all meetings."

# Decision Log

"Create a wiki page at topics/auth-decisions with all authentication-related

decisions from the past month, organized chronologically."

# Code Context

"Search my meetings for any discussion about this function, then use that

context to write better documentation for it."

Example Workflows

Here are end-to-end workflows that show how OneMira C1 fits into a builder's daily routine:

workflows

# Workflow 1: Meeting → Code PR

1. Attend sprint planning (C1 records automatically)

2. Session finalizes → wiki updates with decisions

3. Open Cursor: "What did we decide about the API schema?"

4. Agent searches meetings, finds the schema decision

5. Agent implements the agreed schema with full context

# Workflow 2: Client Call → Requirements Doc

1. Join a client discovery call (C1 captures both sides)

2. After the call: "Summarize the client's requirements"

3. Agent reads session segments, extracts requirements

4. "Write these as a wiki page at projects/client-alpha"

5. Requirements are now searchable and linked forever

# Workflow 3: Weekly Review → Action Items

1. End of week: "List my open action items from this week"

2. Agent reads tasks.md, filters by date and owner

3. "Which ones are overdue? Create follow-up reminders."

4. Context from original meeting decisions is preserved

mira Command Line Interface (CLI) Reference

The OneMira client application installs a powerful CLI utility (mira) to manage your meeting transcripts, search local context, and query your knowledge base directly from the terminal.

Core Commands & Diagnostics

Use these commands to initialize your environment, check system health, and run full-text searches.

mira core

$ mira init

# Initialize Mira's local state directory and wiki vault

$ mira doctor

# Perform self-diagnostics on wiki links, DB integrity, and search indexes

$ mira search "your query"

# Full-text search across transcripts and all wiki documentation

$ mira version

# Print current CLI build version

Session & Transcript Management

Manage your captured meeting records. You can list sessions, display raw transcripts, export to various formats, or manually inject text data.

mira session

$ mira session list --limit 10

# List recent sessions. Append --json to get structured output for scripts

$ mira session show <session-id>

# Output details and raw transcribed dialogue of a specific session

$ mira session export <session-id> --format md

# Export transcript (supported formats: md, txt, srt, json)

$ mira session create --text "Context details" --title "Custom Session"

# Manually create a session (or load from file using --from path/to/file)

$ mira session delete <session-id>

# Permanently delete a session record

Wiki & Knowledge Vault

Query and write directly to your local Markdown knowledge vault files.

mira wiki

$ mira wiki list

# List paths of all wiki pages in the vault

$ mira wiki show topics/pcb-design

# Read the contents of a specific wiki page

$ mira wiki write topics/pcb-design --from doc.md

# Create or update a page from a local markdown file

$ mira wiki backlinks topics/pcb-design

# Find other wiki pages that link back to the target page

$ mira wiki sessions topics/pcb-design

# List session IDs associated with a specific wiki page

Knowledge Base

Once OneMira captures a session, Mira processes the transcript and synthesizes the useful parts into local Wiki files (a Markdown vault) that your AI editors can search, link, and build upon.

Directory Structure & Page Kinds

Your wiki is compiled as a local directory structure representing your knowledge base. Select a folder or file to explore its purpose and view a sample markdown structure:

vault-explorer

$ tree vault/

vault/

├── # project - Tracks concrete goals & milestones

├── # topic - Synthesizes designs & glossary terms

├── # person - Profiles stakeholders & requirements

├── # session - Caches raw transcript logs

├── # index - Global checklist of action items

├── # index - Master catalog of active wiki files

└── # log - Audit trail of LLM ingest transactions

# Click any directory/file link above to view its contents

Double-Bracket Wikilinks & Block Anchors

Pages in the vault are interconnected using standard Markdown and double-bracket references:

Wikilinks ([[Target]]): Cross-link pages by title or alias (e.g. [[projects/miralamp]]) to establish conceptual links.
Block Anchors (^block-id): Use stable anchors at the end of lines to cite specific meeting decisions or quotes directly.
Auto-Aggregated Blocks: Bounded by  comments. The system automatically inserts co-occurring sessions and links during background compilation.

Core Workflows (How to Use)

Integrate OneMira into your daily workflow with three essential tasks:

Ingest (Auto-Update): When a session is finalized, OneMira automatically parses speakers, updates projects, people, and topics, and appends to journal.md.
Query (Ask AI): Ask your AI client (Cursor, Claude Code, etc.) questions about your meetings. It will query the OneMira MCP server to read your vault files, search context, and compile answers citing specific pages.
Lint (Maintain Health): Periodically run health checks to identify ghost links (pointing to nonexistent files), resolve tag drift, and triage reports awaiting promotion.

→

The result is not a raw transcript archive. It is local, structured Context that co-evolves with your work, supporting double-bracket wikilinks ([[target]]), block anchors (^block-id), and automated relational aggregation.

Troubleshooting

Device not found during pairing

Make sure you long-press the side button for 3 seconds to enter pairing mode, then follow the app pairing flow.until the app and device show pairing status. If the device does not respond, ensure it has power. Also ensure Bluetooth is enabled on your computer and keep the device within about 5 meters during pairing and sync..

Bluetooth permission denied

Open your operating system's Bluetooth privacy settings, ensure the OneMira app has permission, then restart the app and try scanning again.

ASR model download fails

Check your internet connection. The model download is typically 100-500 mb depending on the model.. Click "Retry Deploy" to resume. The download will pick up where it left off.

AI tool not detected after setup

Some AI tools need to be restarted after config changes. Close and reopen the AI tool, then call Mira again through its MCP tools, slash commands (such as /mira-search), or tool references (such as @mira_search) depending on your AI client. You can also go to Settings → AI Access → Rescan to verify the injection status.

Device connected but no transcription

Ensure capture is active in the app. Press⌘ R in the app or short-press the side button to start capture. Also check that the transcription engine has been deployed in the setup flow.

How to update firmware

Firmware update details are shown in the desktop app when an update is available..

FAQ

What is OneMira C1?

A local context device. It captures what is said at your desk, turns it into plain-text notes on your own computer, and makes them available to your own AI over MCP. It is not a cloud note-taker—no transcripts leave your machine. Your AI does the thinking; C1 just hands it the spoken context it never had.

How is it different from other cloud AI recorders?

Summaries, recaps, mind maps—every recorder does those. The difference is what happens next. They stop at their own built-in summarizer; C1 hands the conversation to your own agent—Claude, Codex, Cursor—which takes it much further: a real spec, a working doc, a ticket, shipped code. And it is all 100% local, with no subscription and no quota.

Where is my data stored? Does anything go to the cloud?

Everything stays on your own computer as plain-text files. Speech recognition, transcription, and speaker labeling all run locally, and the audio is deleted after it syncs. Nothing is uploaded—and because there is no shared cloud, your conversation can never end up on someone else's account.

Does it need internet or Wi-Fi?

No. C1 has no Wi-Fi—it connects to your computer over Bluetooth or USB-C, and all processing is offline. The only time the internet is involved is when your own AI tools use it.

Which AI tools does OneMira C1 work with? Does it support Chinese dialect transcription?

OneMira C1 works natively with Claude Desktop, Claude Code, Codex, Cursor, and other MCP-compatible or local-file/CLI-based AI tools. Additionally, its local transcription engine supports auto-detection and accurate transcription of English, Mandarin, and major Chinese dialects (like Cantonese and Sichuanese) 100% offline directly on your own machine.

How does my agent get to the captured context?

It pulls it from C1 over MCP. Ask your agent in plain language—it finds and calls the right command and works from the context C1 has stored. C1 is the store; your agent does the work.

Does C1 edit my AGENTS.md or project files?

No. C1 is a passive store—it never reaches into your repo. Your agent, or a scheduled loop you set up, reads from C1, and you decide what gets written into your AGENTS.md, rules, or docs.

Do I need to learn any commands?

No. Just tell your agent what you want in plain language; it discovers and calls the right command on its own. Nothing to memorize.

Can I use it across multiple machines or agents at once?

Yes. C1 is one shared context store—run Claude on one machine and Codex on another, and both connect to the same C1 over MCP and work from the same context.

Can it capture online meetings from Zoom, Microsoft Teams, or Google Meet?

Yes, absolutely. OneMira C1 is designed to capture remote calls and online meetings seamlessly. Even if you are wearing headphones, it captures both your voice and the audio from the far end of the call (requires macOS 14.2+ on Mac for Core Audio process tap), solving the common recording limitations of online meeting platforms.

Will it take over my mic or audio devices?

No. C1 never occupies your computer's mic or audio inputs—your calls and other apps keep working normally while it records.

How good is the audio in a noisy room? Does it tell speakers apart?

C1 uses a dedicated AI chip with hardware noise cancellation to keep voices crisp and clear even in a noisy room, and it labels who is speaking automatically.

How do I start recording?

One press to start, or set scheduled start/stop windows so it is already recording before you remember to.

Can it cover a big conference room?

C1's effective pickup is about 3 m—it is designed as a close-range desk device. It is at its best for desk chats, small-room discussions, and calls; a large hall is not what it is built for.

How is my privacy protected?

By design. C1 is completely offline, all processing happens on your own computer, and audio is deleted after sync. Your voice never leaves your machine.

Do I need consent to record?

It is good practice to let others know before you record—laws vary by place, so check if you're unsure. A simple line works: "Mind if I run my AI notes tool so I don't miss anything? I'll share the notes with you after."

Is it compliant with GDPR, HIPAA, etc.?

Because everything is processed locally with no external transmission, sensitive data never leaves your machine—which makes privacy-heavy work far easier to keep compliant with regimes like GDPR and HIPAA.

Is there a monthly fee or a minute quota?

No subscription, no quota. Capture as much as you want, forever, at zero cost—you just use the AI tools you already pay for.

Battery, storage—how long can it record?

A 1250mAh battery gives up to 30 hours of continuous recording and 60 days of standby. Storage is 32 GB, split between a 15 GB circular buffer for temporary Opus audio files and 17 GB for text snapshots. Because older synced audio is automatically cleared when the 15 GB buffer fills up, it will never run out of space.

What systems does it support?

macOS 12+ and Windows 10+. It connects over Bluetooth 5.4 (LE Audio) or USB-C.

Could my private recordings ever leak to another user's account due to a cloud glitch?

No. Traditional cloud-based AI recorders upload your audio to a shared cloud database, exposing you to the risk of account-mixing bugs or configuration leaks. OneMira C1 is structurally immune to this: everything is processed locally on your own computer, and audio is deleted immediately after syncing. There is no shared cloud database, making cross-user leaks physically impossible.

Why is there no subscription fee or transcription limit? Can I really use it infinitely?

Yes. Cloud-based recorders charge ongoing subscription fees to cover their server hosting and API usage costs. Because OneMira C1 performs speech-to-text and transcription locally on your own machine's hardware, there are no ongoing cloud transcription server bills. We pass this structural advantage to you—a true once-off purchase with unlimited transcription, zero subscriptions, and zero quota anxiety.

Do I have to use a proprietary app or cloud portal to search and organize my data?

No. Traditional recorders lock your transcripts inside their closed database or laggy proprietary app, turning your archive into a "data swamp." OneMira C1 stores all transcripts as simple, clean markdown and plain-text files directly on your local filesystem. You have full ownership: search, grep, edit, and version-control your notes using your preferred editor (like VS Code or Obsidian) and feed them directly to your AI.

How does C1 fit into my "Loop Engineering" workflow?

Your coding agent only knows what is written in your AGENTS.md, system rules, and codebase. But the most valuable rationale—why you rejected an architecture, how a key bug was solved, or what the user actually wants—happens out loud at your desk and never gets written down. C1 captures this spoken layer. You can set up a scheduled shell loop to pull the day's local transcripts and merge the decisions directly into your AGENTS.md and rule files, making your agents starting tomorrow much smarter.

Can I pipe C1 data directly into local CLI tools?

Absolutely. Because OneMira C1 is built CLI-first and outputs standard markdown files, you can pipe them directly into ripgrep, write scripts to parse them, or use our zero-dependency local MCP server (scripts/onemira-mcp.js) to expose them to any LLM-based command-line tool. No closed APIs, no webhooks, no rate limits.

Where do you ship?

OneMira C1 currently ships to addresses within the United States for the first launch. Additional markets are planned later.

What if I decide C1 is not for me?

You can request a return within 30 days of delivery for any reason. Returned items must be in good, functional condition and include all original packaging, cables, accessories, and documentation.

What warranty is included?

OneMira C1 includes a 1-year limited warranty against manufacturing defects from the date of delivery.