remembero
Try the playgroundView on GitHub

Memory you
can reason with.

Store facts and rules as readable knowledge. Ask useful questions. Get deterministic answers with the proof attached.

Try the playgroundView on GitHub
Local-first by default. Logic owns the answer.
Question

Who is collaborating on Atlas?

Querycollaborator(Person, atlas)
Answer

Maya is collaborating on Atlas.

Because
  1. 1project_owner(atlas, rahul)
  2. 2project_contributor(atlas, maya)
Atlas planning session · 17 AugOpen in playground

Not another vector store.

Similarity finds nearby text. Remembero proves what follows.

Readable memory

Plain-text facts, rules, and constraints.

project_owner(atlas, rahul).
status(atlas, blocked).

Deterministic rules

Same knowledge. Same query. Same answer.

needs_follow_up(Person, Project) :-
  promised_update(rahul, Person, Project),
  status(Project, blocked).

Proof, not vibes

Every derived answer carries its supporting claims.

collaborator(maya, atlas)
├─ project_owner(atlas, rahul)
└─ project_contributor(atlas, maya)

The database is the demo.

Insert a SQLite row, run Datalog, then inspect the exact facts and rule behind the answer—all inside your browser.

Open the full playground
Remembero SQLite and Datalog IDE showing tables, a query, proof, and graph
SQLite owns the rows. Ordinary tables remain the storage authority.Rules own the query. The C extension executes inside SQLite WebAssembly.Proof owns the answer. Every result can show its complete support chain.

See what better tools do for a small model.

Three browser workbenches expose the full chain: optional Hermes 7B WebLLM inference with native tool calls, deterministic memory and policy, then SQLite + Wasm execution with the call, result, proof, and timing evidence on screen.

Chat recall labSame small model.
Better tool.

Watch Hermes 7B issue native WebLLM tool calls against one shared SQLite database: raw SQL rows in one lane, Remembero bindings and proof in the other.

Grounded agent labLet the model propose.
Let rules decide.

Run the same Hermes 7B model with and without memory, then watch the request facts, packet swap, gate query, rule, and proof chain stay visible while the action resolves.

SQLite + Datalog playgroundMutate SQLite.
Measure the proof.

Insert real rows, execute the Remembero extension inside SQLite WebAssembly, and inspect the browser-local tables, compiled rule, proof graph, and current-browser timings.

An answer is only useful if you can inspect why.

1

Store evidence

Capture a fact with the statement it came from.

project_owner(atlas, rahul).
2

Apply reviewed rules

Derive useful knowledge without storing invented conclusions.

collaborator(Person, Project) :- …
3

Return the support chain

Inspect the exact claims and rules behind every answer.

answer → rule → sourced facts

Models translate.
Rules decide.

Natural language can translate a question into a query. Remembero evaluates the accepted query against explicit knowledge and returns the evidence locally.

  1. Question natural language
  2. Translate model
  3. Query accepted
  4. Evaluate rules + facts
  5. Answer + evidence

One memory layer.
Three ways in.

MCPConnect agents and tools through Model Context Protocol servers.
TypeScriptUse the typed library API inside your applications.
CLInpx -y remembero

Build agents that can show their work.

Try a real-life lab first, then open the IDE when you want to inspect the machinery.

Open a labOpen the playground