Emem

What Vortx-AI/emem does

Vortx-AI/emem gives agents a common way to identify places, real-world objects, and signed observations. A named location can resolve to a canonical cell64, while an object such as a bridge, farm plot, river, or named place can receive an emem:entity: token. Agents can pass these handles between conversations or peers without rewriting the original description.

The service exposes Earth-related facts including elevation, land cover, surface water, vegetation, and built-up context. Facts carry signed receipts and stable identifiers. A memory token points to a signed fact, while a bundle token can cover up to 256 cell, band, and time combinations in one citation envelope.

The public read endpoint requires no account or API key. The system separates published commitments from source files: a signed record can identify and verify selected data without exposing unpublished bytes.

How it works

A typical workflow starts with emem_locate, which resolves a place name or coordinates and reports available data bands and algorithms. emem_recall then reads facts for that cell and band, fetching and signing a value when a registered materializer exists. Initial materialization can be slower; later reads use the cache.

For a single packaged response, emem_ask routes a place-based question through location, recall, and algorithm steps. emem_intent provides a typed alternative for goals such as finding a place, checking what is present, measuring change, comparing cells, or confirming a claim.

Citations can be created with emem_memory_token or emem_memory_bundle and resolved later with emem_memory_token_resolve. emem_verify_receipt checks the receipt byte for byte, while emem_echo_verify compares an emitted value with the signed value to catch transcription drift. emem_memory_contradictions exposes disagreement between attesters instead of selecting one value silently.

Setup and configuration

Connect an MCP host to the streamable HTTP endpoint:

claude mcp add --transport http emem https://emem.dev/mcp

The service can also be used through the published Python and TypeScript packages. Python examples use ememdev[signing], and the TypeScript package is @vortxai/emem. Offline receipt verification is available through the Python signing utilities and uses the public key embedded in the receipt. To pin the service identity, retrieve the key from /.well-known/emem.json.

No environment variables or API keys are needed for public reads. MCP clients should call tools by their MCP names rather than guessing REST paths; the OpenAPI document defines the HTTP API paths.

Tools and capabilities

• Resolve places to cell64 addresses with emem_locate.
• Mint, find, and link shared object identities with emem_entity, emem_entity_resolve, and emem_entity_link.
• Recall, search, and dereference signed facts with emem_recall, search, and fetch.
• Create and resolve portable citations with memory tokens and bundles.
• Verify receipts, emitted values, provenance, and policy conditions.
• Ask packaged geographic questions or run typed intents for change, similarity, confirmation, and area-over-time analysis.
• Discover the catalog through emem_tools; the core MCP page exposes 18 tools, while the full catalog contains 116.