ChatGPT doesn't know what Claude said. Codex can't see your Cursor session. Orchestra gathers chats, code, and design work into one memory you own — browser, IDE, and desktop agents all read the same database.
No inference markup. Your API keys stay in your environment — never in our database.
You ask ChatGPT for code, Claude for a review, Gemini for research — then manually shuttle context between them.
Each tool only remembers inside its own walls. Last month's breakthrough is trapped in an old chat thread.
One memory layer. Every conversation, note, and file indexed together — searchable, linkable, yours.
How it works
You keep using the tools you love. Orchestra quietly saves what matters and connects the dots — without making you switch apps or learn a new chat interface.
Browser extension for web chats and v0. IDE extension for Cursor. MCP for Claude Code and agents.
Everything lands in one database on your laptop — or Postgres you control in the cloud.
Find by keywords, meaning, or both. Filter by which AI, date, or type of content.
Ask any model with full context. Run consensus or a four-step review pipeline on past work.
Works everywhere
Web chats and desktop coding tools don't share memory by default. Orchestra bridges them — same search, same recall, whether you're in a browser tab or inside Cursor.
Auto-sync web chats, Recall related memory, and Insert into chat from the Memory panel.
Search memory, recall context, save selections — while you code.
Let the AI call memory tools directly — no copy-paste.
memory_context, memory_save, searchAll three surfaces write to the same Orchestra memory
Smart search
Orchestra combines classic keyword search with meaning-based search. Ask “that postcode regex from last month” even if you never used the word “regex.”
When you search, Orchestra blends two approaches:
Every search result shows which AI it came from and when. One click opens the original ChatGPT or Claude thread in your browser.
Team of AIs
Orchestra can ask several models the same question and show how much they agree — or run a structured pipeline where each step sees what the last one produced.
Same question to Claude, GPT, Gemini, and others. Get an agreement score plus one synthesized answer.
Pipeline and chat modes pull relevant memory automatically — no manual paste.
See it
The desktop shell wraps the same web UI. The browser extension captures chats and injects recall. Cursor and VS Code use the IDE extension plus MCP for agents.
Chat, multi-agent pipeline, consensus, hybrid search, history, and Setup (export/import).
Capture on Claude Code, ChatGPT, and 40+ sites. Memory panel inserts recall into the composer.
Search memory, recall context, save selections — same database as the browser bridge.
Everything included
A coordination platform — not another chat window.
Chats, Codex, v0, Claude Code — Save, Recall, Memory panel with Insert into chat. Tier A sites get tailored extractors.
Cursor, VS Code, Windsurf: search memory, Recall (Ctrl+Shift+R), save selections with source labels.
Cursor Agent, Claude Code, Claude Desktop — memory_context, save, hybrid search as tools.
Keywords + meaning-based search. Filter by Cursor, Claude Code, Codex, or any web source.
Index local code and design docs on startup — project files become searchable memory.
SQLite locally or Postgres you control. One DB for browser, IDE, and MCP.
Portable JSON backups. CLI export / import bundle. Move laptop → cloud.
OpenAI, Anthropic, Gemini, DeepSeek, Grok, OpenRouter, Ollama, LM Studio. No inference markup.
Full web app, CLI, HTTP API, optional Tauri desktop, Setup tab for export/import.
Your data
Same product — you pick where the memory file lives.
A single file on your machine. Nothing leaves your PC unless you export or sync folders you choose. Perfect for solo work and air-gapped setups.
Orchestra runs on a small server. Your chats live in Postgres you own — Neon, Supabase, or your own host. We run the app; you hold the keys.
Pricing
Pay for Orchestra — not inflated inference. LLM bills go directly to the providers you already use.
+ your own API keys (pay providers directly)
+ Postgres (~$7–25/mo at Neon, Supabase, etc.) + API keys
Infrastructure pass-through included in price
Compare
Most tools are great at talking — Orchestra is built for remembering and connecting across tools.
| Orchestra | ChatGPT / Claude alone | Built-in AI memory | Mem0 / memory APIs | Notion AI | |
|---|---|---|---|---|---|
| Memory across multiple AIs | Yes — all in one place | No — one product only | Per vendor, not shared | Dev-focused, not end-user UI | Notes only, not chats |
| Capture browser chats automatically | 40+ sites + extension | Inside that app only | Limited | Via API integration | No |
| Codex, v0, Figma, code builders | Browser extension | Per product | No unified memory | Build yourself | Notes only |
| Cursor / VS Code integration | IDE extension + MCP | Separate products | Varies | Dev API only | No |
| Claude Code CLI memory | MCP tools | No | Claude-only | Custom wiring | No |
| Search all past AI work | Keywords + meaning | Within one account | Within one account | Search API | Workspace search |
| Ask multiple models one question | Consensus mode | Switch apps manually | No | No | No |
| Multi-step AI workflow | Architect → Tester pipeline | Manual | No | Build yourself | No |
| You own the database | Local or your Postgres | Vendor cloud | Vendor cloud | Depends on setup | Notion cloud |
| Bring your own API keys | No inference markup | Pay vendor only | Pay vendor only | Plus their fees | Bundled subscription |
| Export full memory backup | JSON bundle | Limited exports | No | API-dependent | Export notes |
| Open thread in original AI | One click from search | Native | Native | No | No |
| Typical cost (beyond LLM usage) | $59 once or $12–19/mo | $20–200/mo subscriptions | Included in sub | Dev tiers + hosting | $10–20/mo per seat |
Questions
Desktop terminals use the IDE extension (Cursor/VS Code) or MCP (Claude Code CLI). The browser extension covers Claude Code on claude.ai/code, Codex, v0, and web chats. All three share one database.
Install extensions/vscode, enable MCP in .cursor/mcp.json, and run npm run dev. Use Recall (Ctrl+Shift+R) or Agent tools like memory_context.
No. You keep using them. Orchestra remembers what they said so you — and any other model — can use it later.
On your computer by default (a local file). Cloud plans let you point to Postgres you control — Neon, Supabase, Railway, or your own server.
On the BYO-database plan, memory stays in your Postgres. We only run the app layer; your connection string is in your environment.
You pay OpenAI, Anthropic, and others directly. Orchestra doesn't resell tokens or add a markup.
Yes. Export a JSON bundle from Setup (or orchestra export), deploy Orchestra with Postgres, and import the bundle. Conversation IDs, messages, and artifacts are preserved — same memory, new home.
Start local in minutes. Scale to cloud when you need remote access or a team database.
Choose your plan