Getting started
Pick up where you left off.
You explained your project, chose a direction, and assigned the work. Tomorrow, you open a new conversation. How does your assistant know what you decided?
ConceptMem is a managed database that keeps shared current state your assistants can read and propose updates to. When a decision changes, it consolidates the new information under your database rules and preserves the history. You can return to the current record instead of explaining the project again.
Let’s follow Project Aurora from its first saved decision to a new conversation. You’ll create a database, connect your assistant, and ask it to remember who is responsible and how the project is progressing.
1. Create your account and database
Sign up for ConceptMem, complete the account flow, and sign in. Already registered? Sign in to the platform to continue.
Open your databases and select Create database. Name it Project Aurora. A database holds the context you want to share: a project, a team, or your personal work. In connection settings, you may also see it called a workspace.
Before creating it, choose its schema governance using the options below.
2. Choose how its structure evolves
Your database’s ontology describes the kinds of things it can remember, their fields, and their relationships. For Aurora, that might mean projects, people, project status, and who works on what.
| Mode | How it works | When to use it |
|---|---|---|
| Flexible | Your agent can build and evolve the ontology as you go, with vocabulary and consistency checks. | You are discovering what you need to remember. |
| Model-driven · Semi-restricted | Start from a base metamodel. The agent can propose changes to the structure that pass the model’s validation rules. | You want a shared starting structure that your agent can extend. |
| Model-driven · Restricted | Start from a base metamodel. Agents can propose observations; a human editor manages classes and fields in the platform. | You want people to control changes to the structure. |
For this walkthrough, choose Flexible and keep Include general-purpose starter classes for people, places, events, goals, and work selected. Then select Create database. For either model-driven option, choose a Base metamodel and select Semi-restricted or Restricted under AI changes to the model.
3. Connect your assistant
Open Connections in your database and choose the MCP integration. MCP is the protocol that lets an assistant use ConceptMem’s tools from a conversation. Copy the endpoint shown for Project Aurora and follow the setup for your client.
Authorize with your account (OAuth)
- Add the database’s MCP endpoint to a client that supports remote MCP with OAuth.
- Start its sign-in flow and sign in to ConceptMem in the browser.
- Review the application, select the database, and choose the access level. For this walkthrough, choose Read and write if your role allows it.
- Allow access, return to your assistant, and enable the connection.
OAuth lets you authorize the application without creating an API key. Read-only access works for retrieving context; saving the examples below requires write access.
Connect with a token (API key)
- In Connections, use the API key setup and select Create API key.
- Save the generated key before leaving or reloading the page; its secret is shown once.
- Copy the configuration into your client’s MCP settings, including the endpoint and key.
- Enable the connection and confirm that the ConceptMem tools are available.
Keep the key in your client’s authentication configuration. Manage keys and connected applications in the database’s Settings. If you cannot create a key, ask a database owner to review your role, or use OAuth for read-only access.
Connections includes setup tabs for ChatGPT, Claude Desktop, Cursor, and Codex. For another client, such as Claude Code, use its remote MCP setup with the same database endpoint and a supported authentication method. The Connection guide explains both paths.
Use ConceptMem to list the classes and fields available in the connected database.The assistant should consult ConceptMem and report the available vocabulary. If it cannot find the tools, check that the MCP connection is enabled in this conversation before continuing. Mentioning ConceptMem in a message does not connect it by itself.
4. Save and check your first information
Now give your assistant something you want to carry into the next conversation:
Use ConceptMem to record that Ana is responsible for Project Aurora
and the project is in planning. Reuse the existing vocabulary and
record this conversation as the source. Tell me what was saved
and whether anything was skipped, refused, or needs review.Your assistant sends a proposal. ConceptMem evaluates it against the database rules and returns what happened. Check that result: a completed request does not mean every field was accepted. If a proposal needs review, resolve it in the database’s review area before treating it as current information.
Consult ConceptMem: who is responsible for Project Aurora, and what is its current status?If both details were accepted, the answer should identify Ana and the planning status. You can also open the database’s entities in the platform to inspect the saved record.
5. Continue in another conversation
Open a new conversation with the assistant connected to the same database. You can also use another assistant after configuring and authorizing its access to Project Aurora.
Use ConceptMem to recover the current context of Project Aurora before we continue.The assistant can now retrieve the saved project details. This shares the information recorded in the database; it does not automatically transfer your entire chat history. When the project moves forward, say what changed:
Project Aurora has moved from planning to execution. Update the same
project in ConceptMem, then check its current status and the history
of that change. Tell me whether the update was applied.Under the walkthrough’s default policy, an accepted update makes execution the current status and preserves planning in history. The recorded source and time help you trace the change. If your policies hold or refuse it, the assistant should explain that outcome.
6. Make it part of your routine
You can explicitly ask to use ConceptMem whenever you want to save or retrieve context. For recurring use, add the following text to your assistant’s persistent instructions or project instructions, replacing Project Aurora with your database’s name:
Use the connected ConceptMem database for Project Aurora as shared governed state.
Before answering questions that depend on earlier decisions, owners,
preferences, or progress, read current facts from it.
When I communicate a relevant decision or change, propose an update
with its source. Reuse existing records and vocabulary.
Check every write result and tell me whether it was applied, skipped,
refused, or held for review. Do not record assumptions as facts.
If the target database is unclear, ask me which one to use.
If ConceptMem is unavailable, say so; do not claim information was saved.Save these instructions wherever your client supports ongoing guidance. They guide the assistant’s behavior; the MCP connection must still be configured, authorized, and enabled in each application you use.
Your next step
Choose one real decision, ask your assistant to save it, and retrieve it in a new conversation. That is the everyday cycle: remember what matters, recover the context, and update what changed.
Building an integration?
Use these references for tool arguments, application code, and write results.
MCP
Connect a client, use the tools, and read the results.
Read the guide →Connections
Authorize agents and applications through OAuth or an API key.
Read the guide →Write, read, and history
Run a complete two-source customer example with explicit schema and policy.
Read the guide →Members
Invite teammates and manage roles, keys, and application access.
Read the guide →Glossary
Understand current state, proposals, and governance modes.
Read the guide →TypeScript SDK
Configure the client and work with current state and history.
Read the guide →Public types
Understand observations, records, provenance, and write results.
Read the guide →