Carry a project convention into your next agent session

Start with one small piece of project knowledge: a package-manager choice, a test command, or a naming convention. Store it once, open a fresh agent session, and inspect what comes back. This exercises the free local core; a Pro account is not required.

The commands below follow the published v1.7.8 installation documentation and agent connection guide. A documented configuration recipe is not a claim that every client version has been certified.

1. Install the local service and connect one agent

The full installation requires Python 3.10 or later. Run these commands in the Python environment you want your agent and dashboard to use:

The dashboard opens at http://127.0.0.1:8700. Keep it local. Configure your agent's MCP client to start engraphis-mcp from that same installation. For example, the published Claude Code recipe is:

For another client, follow its current MCP configuration instructions and the published connection guide. Reopen the client if its setup flow requires it, then confirm it can see the Engraphis tools.

2. Store one convention with an explicit scope

Use a harmless convention from a project you control. The names below are examples; substitute your own workspace and repository consistently in both sessions.

Use Engraphis to remember this convention in workspace "memory-guide", repository "example-app": frontend packages use pnpm. Keep it scoped to this repository.

Check the write result and open the corresponding memory in the dashboard. Confirm the stored text and repository scope. Keep credentials, personal information, and temporary scratch notes out of this exercise.

3. Ask from a fresh session

Open a new agent conversation in the same repository, using the same local Engraphis installation. Ask the question without repeating the answer:

Which package manager should I use for frontend packages? Check Engraphis memory in workspace "memory-guide", repository "example-app", and show the source and scope of the answer.

Look for an actual recall result and its supporting memory. An assistant repeating the correct sentence is not enough on its own: inspect the record, boundary, and recall rationale in the dashboard.

4. If the result is missing, check the connection first

• Confirm the client can call the configured Engraphis MCP server.
• Check that both sessions use the same workspace, repository, and local database.
• Verify that the original write succeeded and that the memory is available for recall.
• Run engraphis-init --check --json for setup diagnostics. Read the result before changing configuration.

This is a workflow check, not a retrieval benchmark or a guarantee that every question will return the expected fact. Try a real convention you need later, then inspect why it was or was not recalled.

When Pro becomes useful

Once you have a workspace with useful history, Pro adds Analytics, scheduled Auto Consolidation and Auto Dreaming, and optional Cloud Sync. Use those tools to review memory health and choose maintenance, rather than treating a successful local recall as proof of a hosted result.

Read what stays local and what you choose to send, then follow the Pro trial checklist. A new workspace may need more history before Analytics can show a useful trend.