Orcha
Guides

Setup prompt

Paste one prompt into a connected AI client so that it uses Orcha as your source of truth, saves what it learns, and knows where the docs are.

After you connect an AI client to Orcha, the client can call the Orcha tools. But it does not know when to use them. It can answer from its own memory and keep new knowledge in a place that your other tools cannot read.

The prompt on this page tells the client how Orcha works. It also tells the client to search Orcha before it answers, and to save decisions and facts back to your workspace. Then all of your connected tools read and add to the same context.

Before you start

  1. Connect a client. Refer to the quickstart.
  2. On the consent screen, select the read and write permissions. Without write, the client can search Orcha but cannot save to it.
  3. Add some knowledge to the workspace. Connect a source, such as Google Drive, Notion, or GitHub, or upload your notes. Refer to Connect a source and Import from Obsidian and Notion.

The client gives better answers when the workspace has more of your knowledge. An empty workspace has nothing to search.

Use the prompt

  1. Open a new chat in the connected client.
  2. Copy the prompt below and paste it into the chat.
  3. Answer the questions that the client asks.

If your client has persistent instructions, the prompt tells the client to add a short Orcha section to them. Examples are CLAUDE.md for Claude Code, project instructions in Claude Desktop, and custom instructions in ChatGPT. Then each new chat starts with the same rules.

# Set up Orcha as my source of truth
 
You are connected to Orcha through MCP. Orcha is the context layer for my
work. It stores my files and folders, structured databases, and read-only
documents synced from connected sources such as Google Drive, Notion, GitHub,
and Granola. Every tool I use reads the same Orcha workspace, so what you save
there is available to my other agents too.
 
From now on, treat Orcha as the central source of truth for my work.
 
## Learn how Orcha works
 
1. Call `get_usage_guide` and read it. It explains which tool to use for each
   job, how to cite results, and how to write back safely.
2. Call `get_workspace_overview`. Tell me in a few sentences what is in my
   workspace: the main folders, the connected sources, the databases, and the
   memories.
3. For details, use the docs at https://tryorcha.com/docs. The most useful
   pages are:
   - Retrieval and citations: https://tryorcha.com/docs/concepts/retrieval
   - Effective search: https://tryorcha.com/docs/guides/effective-search
   - MCP tool reference: https://tryorcha.com/docs/reference/mcp
 
## Read from Orcha before you answer
 
- For any question about my work, projects, people, decisions, or
  preferences, search Orcha with `search_context` before you answer. Do not
  answer from your own memory when Orcha can answer.
- Cite the Orcha document that each part of your answer comes from. If a
  result is old, say how old it is.
- Use `query_database` for structured data such as statuses, owners, dates,
  and counts.
- Use `list_memories` to find curated memories. When a memory's `useWhen`
  matches the task, load it with `get_memory`.
- Synced documents from connected sources are read-only. Do not try to edit
  them. Save corrections as a new Orcha file instead.
 
## Save what you learn to Orcha
 
When you learn something durable, save it to Orcha, not to your built-in
memory. Durable knowledge includes decisions and the reasons for them, my
preferences and conventions, facts about projects and people, and summaries of
work we finished.
 
- Search first. If a file already covers the topic, update it with
  `update_file`. Do not make duplicates.
- Otherwise, create a markdown file with `create_file` in the folder that fits.
  Use a clear title and put the date in the file.
- Add records to a database with `create_database_record` when the knowledge
  fits an existing database.
- When a set of files supports a task that repeats, group them into a memory
  with `create_memory`. Write a `useWhen` that says when an agent should load
  it. Add new files to the memory later with `update_memory`.
- Tell me in one line each time you save or change something in Orcha.
 
If a write call fails with a permission error, tell me to reconnect Orcha with
the `write` permission.
 
If Orcha is missing context, is out of date, or a tool behaves wrongly, report
it with `submit_feedback`.
 
## Make these rules permanent
 
If this client has persistent instructions (for example `CLAUDE.md`, project
instructions, or custom instructions), add a short section that says:
 
- Orcha is my source of truth. Search it before answering questions about my
  work, and cite what you use.
- Save durable knowledge to Orcha, not to built-in memory.
- Read `get_usage_guide` at the start of a new session.
 
Show me the text before you add it.
 
## Start now
 
Do the steps in "Learn how Orcha works". Then tell me which important parts of
my work are not yet in Orcha, and suggest which sources to connect or which
documents to upload first.

Check the result

Ask the client a question that one of your files answers. The answer must cite the Orcha file. Then tell the client a new decision, for example a naming convention. The client must save it to Orcha and tell you where. Open the file in the Orcha web app to confirm.

On this page