Skip to content

Conversational setup

Verified for MemStrata Conversational 0.7.19 on 2026-10-12. Coder editor plugins and token-savings pages belong to Coder, not this product.

  • Python package (verified): Download Conversational. Select Windows x64, Linux x86_64, or macOS 15 Apple silicon. Copy the full command. It creates a dedicated environment, installs the 0.7.19 wheel, checks --pypi-package-version, then runs init.
  • Microsoft Store: a listing can exist while a newer submission is in certification. The wheel command is the verified current installer.
  • Short pip (already in a Python 3.10+ environment):
Terminal window
python -m pip install --upgrade --only-binary=:all: memstrata-conversational==0.7.19
python -m memstrata_conversational_launcher init

On Windows, if you used the guided command, the launcher is memstrata-conversational.cmd.

Start the product from the launcher printed at the end of install. Complete sign-in and activation. The 180-day trial starts after successful activation, not after download. Reinstalling does not restart it.

If activation fails, see service information and email support@memstrata.dev.

Conversational does not ship an answering model. Connect the AI you already use. MemStrata selects evidence from the local index and shows the passages with the answer.

Optional visual retrieval selects pages. It does not write the answer. Platform and evidence limits: Platform and model notes.

Download the tiny demo file: project-notes.md. It contains two fictional notes: a Tuesday review on 10 March and a Friday correction on 12 March.

Import that file (or one note you already know). Wait until indexing finishes and the source shows Ready. Do not ask while the source is still indexing.

If indexing fails: check that the file is a supported text note, not an unsupported chat-export schema. Retry with this demo file first.

Select the demo source. Ask:

When is the project review?

The expected answer names Friday.

Open Why this answer (or the equivalent evidence control in the installed UI). You should see the Friday correction as the supporting record. A cited source makes the answer easier to check. It is not a guarantee that every answer is correct.

The 0.7.13 screenshots of Setup, Chat (Standard 8k / Deep 24k), and Why this answer remain on the product walkthrough.

After the first answer works, switch evidence mode only if the installed UI offers Standard and Deep.

ModeEvidence limitWhen to use it
Standard8,192 tokensDefault
Deep24,000 tokensWhen Standard evidence is not enough

Those limits are retrieved-source budgets, not the model’s full context window. The R454 research run used a 7,000-token evidence budget. That is not the product Standard setting.

Optional ColQwen2 and ColFlor select original pages. They are not the answering model. EVIE and ColPali are coming soon and are not listed as supported downloads. Detecting a GPU is separate from installing a compatible runtime.

Conversational imports chats, notes and documents. It does not replace Coder editor plugins. Browser-extension and coding-host pages belong to Coder.

If the port is in use: close the other MemStrata product using the same loopback port, or follow the on-screen port message.

If your model is not connected: connect the AI you already use, then ask again. MemStrata does not include an answering model.

If evidence is empty: wait until the source shows Ready, then ask again. Inspect the evidence panel before treating the answer as complete.

If activation fails: check status and email support@memstrata.dev.

To move from 0.7.18 to 0.7.19, run the current Download command again in the same dedicated environment, or python -m pip install --upgrade --only-binary=:all: memstrata-conversational==0.7.19. Then confirm --pypi-package-version prints 0.7.19.

Follow Uninstall for the Conversational environment you created. Do not uninstall a different product’s environment by mistake.