Dictate anywhere.
Your voice becomes clean text.
Hold fn in any app. Vordi transcribes, cleans, and types into the focused field while the notch shows exactly what is happening.
Vordi turns everything you dictate into a private, on-device memory. Then Claude, Cursor, and Codex read it through one local MCP.
Recommended
Vordi in action
The loop below runs continuously: connect, search local transcript history, resolve context, then return an answer with its sources.
Resolved 3 relevant transcripts and ranked the onboarding decision with its original context.
You decided to request permissions progressively, then move API-key setup after the first successful dictation.
request permissions progressively
move API keys after activation
explain every permission in context
You decided to request permissions progressively, then move API-key setup after the first successful dictation.
Real product proof
A complete Vordi workflow, from dictation across apps to an agent retrieving the context later.
Four parts, one memory loop
Hold fn in any app. Vordi transcribes, cleans, and types into the focused field while the notch shows exactly what is happening.
Run history lives under Application Support. Local SQLite, FTS5, embeddings, and entity links turn past dictations into useful memory.
Vordi bundles a dependency-free, read-only MCP server and provides an in-app connection flow for supported local agents.
Search returns ranked transcript excerpts and source metadata, so the response can point back to the underlying run instead of guessing.
You decided to request permissions progressively and defer API-key setup until after the first successful dictation.
Dictate and run commands into the focused field of any app. No plugins, no integrations.
Ask what you forgot
Questions stay grounded in your own run history, with source snippets the agent can cite.
“What did I decide about onboarding?”
“Which launch risks did I call out?”
“What did I promise to follow up on?”
“Find what I said about the pricing plan.”
“Summarize my notes about Project Atlas.”
“When did I mention this regression?”
Connect once. Every compatible agent reads from the same local memory surface.
"vordi": {
"command": "vordi-mcp",
"tools": [
"vordi_search_transcripts",
"vordi_get_run"
]
}Search returns the decision, its timestamp, and the original run instead of an ungrounded summary.
Claude Code, Cursor, Codex, and compatible MCP clients share one memory without duplicating it.
Transcript files, exact search, semantic retrieval, and source relationships stay navigable as one local system.
Field guides
Honest comparisons and practical notes on voice, local memory, and AI-agent workflows.
VordiCompare price, privacy, local memory, provider choice, and what Vordi adds for Mac users and AI agents.
Read the comparisonInstall Vordi
macOS 13 Ventura or newer. Current public releases are ad-hoc signed beta builds and require Microphone, Accessibility, and Input Monitoring permissions.
Homebrew cask
Manual installation