emem, the verifiable memory protocol for the physical world changelog

version

2.4.4 → 2.4.5

https://emem.dev/llms.txt

− Every patch of ground has a 64-bit address (cell64, ~9.55 m at the equator). A fact is keyed by Cell × Band × Tslot and signed by the responder over the BLAKE3 of its canonical CBOR, so the same content id returns byte-identical bytes from any conformant responder and any client verifies the receipt offline. Reads need no auth. For a single-shot answer to a free-text question, call emem_ask: it routes the question to a place, recalls the relevant bands, and runs the applicable algorithms in one − - [algorithms](https://emem.dev/v1/algorithms): 169 composition recipes (flood_risk, walkability, eudr_compliance, …).+ Every patch of ground has a 64-bit address (cell64, ~9.55 m at the equator). A fact is keyed by Cell × Band × Tslot and signed by the responder over the BLAKE3 of its canonical CBOR, so the same content id returns byte-identical bytes from any conformant responder and any client verifies the receipt offline. Reads need no auth. For a single-shot answer to a free-text question, call emem_ask: it routes the question to a place, recalls the relevant bands, and runs the applicable algorithms in one + - [algorithms](https://emem.dev/v1/algorithms): 172 composition recipes (flood_risk, walkability, eudr_compliance, …).

https://emem.dev/llms.txt

− Every patch of ground has a 64-bit address (cell64, ~9.55 m at the equator). A fact is keyed by Cell × Band × Tslot and signed by the responder over the BLAKE3 of its canonical CBOR, so the same content id returns byte-identical bytes from any conformant responder and any client verifies the receipt offline. Reads need no auth. For a single-shot answer to a free-text question, call emem_ask: it routes the question to a place, recalls the relevant bands, and runs the applicable algorithms in one − - [MCP endpoint](https://emem.dev/mcp): JSON-RPC 2.0 over Streamable HTTP; tools/list returns the 18 core tools by default, tier "all" or the /mcp/full endpoint returns all 115. Every tool is callable by name from either endpoint, so narrowing discovery removes no capability; call emem_tools for the map or one tool's schema. Each tool declares one shape in MCP-standard _meta as dev.emem/shape (the form of the answer: scalar, timeseries, raster, geometry, vector, identity, token, proof, plan, fil− - [ask](https://emem.dev/v1/ask): POST free-text question (+ place) → topic-route → recall → applicable algorithms, in one call.− - [algorithms](https://emem.dev/v1/algorithms): 168 composition recipes (flood_risk, walkability, eudr_compliance, …).+ Every patch of ground has a 64-bit address (cell64, ~9.55 m at the equator). A fact is keyed by Cell × Band × Tslot and signed by the responder over the BLAKE3 of its canonical CBOR, so the same content id returns byte-identical bytes from any conformant responder and any client verifies the receipt offline. Reads need no auth. For a single-shot answer to a free-text question, call emem_ask: it routes the question to a place, recalls the relevant bands, and runs the applicable algorithms in one + - [MCP endpoint](https://emem.dev/mcp): JSON-RPC 2.0 over Streamable HTTP; tools/list returns the 18 core tools by default, tier "all" or the /mcp/full endpoint returns all 116. Every tool is callable by name from either endpoint, so narrowing discovery removes no capability; call emem_tools for the map or one tool's schema. Each tool declares one shape in MCP-standard _meta as dev.emem/shape (the form of the answer: scalar, timeseries, raster, geometry, vector, identity, token, proof, plan, fil+ - [ask](https://emem.dev/v1/ask): POST free-text question (+ place) → topic-route → recall → applicable algorithms, in one call. When vouched research agents concluded something inside the place, the answer adds `research_nearby`: pointers only, never their values.+ - [inferences](https://emem.dev/v1/inferences): POST cell or lat+lng (+ radius_m, default 5000) → signed derivations other agents registered near that place, nearest first, each with its parent tokens, attester, tier and distance. The response states its trust policy (default: keys an organisation vouched for). A signature says who concluded something, never that it is true.+ - [algorithms](https://emem.dev/v1/algorithms): 169 composition recipes (flood_risk, walkability, eudr_compliance, …).

https://raw.githubusercontent.com/vortx-ai/emem/HEAD/AGENTS.md

− Two hard-won specifics. CI pins `dtolnay/rust-toolchain@stable`, so the− | MCP tool registry (115 tools) | `crates/emem-mcp/src/lib.rs` |+ Every action in `.github/workflows/` is pinned to a full commit SHA with+ its tag in a trailing comment (`@<sha> # v4.4.0`). A tag can be moved to+ other code; a SHA cannot. To bump one, resolve the new tag with+ `git ls-remote --tags https://github.com/<owner>/<action>` (take the+ peeled `^{}` line for an annotated tag) and change the SHA and the comment+ together.+ Two hard-won specifics. CI installs Rust from the `stable` branch of+ `dtolnay/rust-toolchain` (pinned to a commit like every other action;+ that commit installs whatever stable is current when the job runs), so the+ | MCP tool registry (116 tools) | `crates/emem-mcp/src/lib.rs` |

https://raw.githubusercontent.com/vortx-ai/emem/HEAD/AGENTS.md

− Rust workspace, 19 crates, version 2.4.3, MSRV 1.91. `emem-guard` is the verdict server for AI inference checkpoints (Anthropic Inference hooks, Claude Code hooks); it is a separate binary and shares no code path with the responder. The bulk of the code− `crates/emem-fetch/src/*.rs` (27 data connectors + 7 utility modules).− # Production binary (what /home/ubuntu/.config/systemd/user/emem-server.service runs)− # After release rebuild on systems that need port 443 binding:− sudo setcap 'cap_net_bind_service=+ep' target/release/emem-server− systemctl --user restart emem-server.service− `scripts/redeploy.sh` does build + setcap + restart in one shot.− Crate-level integration tests live under `crates/<crate>/tests/`. Only two− crates have those today: `emem-fact` (round-trip) and `emem-fetch` (live COG fetch).− Both gates run in `.github/workflows/ci.yml` against Linux + macOS, plus− an MSRV 1.91 build job. Don't ship a commit that fails either: the CI− - **No `unsafe`.** Every crate carries `#![forbid(unsafe_code)]` at the top.− | Open-data connectors | `crates/emem-fetch/src/*.rs` (27 data modules + 7 utility) |− | Registries (8 manifests) | `crates/emem-core/data/*.json` + `src/` |+ Rust workspace, 19 crates, version 2.4.4, MSRV 1.91. `emem-guard` is the verdict server for AI inference checkpoints (Anthropic Inference hooks, Claude Code hooks); it is a separate binary and shares no code path with the responder. The bulk of the code+ `crates/emem-fetch/src/*.rs` (27 data connectors + 8 utility modules).+ # Release binary, for a local or self-hosted node+ Production (emem.dev) does not run a binary built on the host. It runs a+ container under the system unit `/etc/systemd/system/emem-server.service`,+ which starts the image with `docker run`. To deploy:+ 1. Push the commit to `main`.+ 2. Wait for the `publish` workflow (`.github/workflows/publish.yml`) to+    finish with success. It builds and pushes the image.+ 3. Run `/home/ubuntu/emem-ops/pending_deploy.sh <new7> <good7>`. Both+    arguments are 7-character commit shas: the new one, and the last good+    one to roll back to.+ Do not use `scripts/redeploy.sh` on the production host. It builds on the+ host and restarts a user unit that production no longer uses. It is only+ for a node that runs the binary directly.+ Crate-level integration tests live under `crates/<crate>/tests/`. Five+ crates have those today: `emem-airgap`, `emem-fact` (round-trip),+ `emem-fetch` (live COG fetch), `emem-primitives` and `emem-trace`.+ Both gates run in `.github/workflows/ci.yml` on Linux. Only the+ `test` job runs on Linux and macOS. There is also an MSRV 1.91 build+ job. Don't ship a commit that fails either: the CI+ - **No `unsafe`.** Every crate carries `#![forbid(unsafe_code)]` at the top,+   except three that do not yet: `emem-airgap`, `emem-scorecard` and+   `emem-sleep-agent`. They use no `unsafe`; add the attribute when you+   next edit them.+ | Open-data connectors | `crates/emem-fetch/src/*.rs` (27 data modules + 8 utility) |+ | Registries (10 manifest digests, `GET /v1/manifests`) | `crates/emem-core/data/*.json` + `src/` |

version

2.4.3 → 2.4.4

https://emem.dev/llms.txt

− Every patch of ground has a 64-bit address (cell64, ~9.55 m at the equator). A fact is keyed by Cell × Band × Tslot and signed by the responder over the BLAKE3 of its canonical CBOR, so the same content id returns byte-identical bytes from any conformant responder and any client verifies the receipt offline. Reads need no auth. For a single-shot answer to a free-text question, call emem_ask: it routes the question to a place, recalls the relevant bands, and runs the applicable algorithms in one − - [MCP endpoint](https://emem.dev/mcp): JSON-RPC 2.0 over Streamable HTTP; tools/list returns the 18 core tools by default, tier "all" or the /mcp/full endpoint returns all 114. Every tool is callable by name from either endpoint, so narrowing discovery removes no capability; call emem_tools for the map or one tool's schema. Each tool declares one shape in MCP-standard _meta as dev.emem/shape (the form of the answer: scalar, timeseries, raster, geometry, vector, identity, token, proof, plan, fil+ Every patch of ground has a 64-bit address (cell64, ~9.55 m at the equator). A fact is keyed by Cell × Band × Tslot and signed by the responder over the BLAKE3 of its canonical CBOR, so the same content id returns byte-identical bytes from any conformant responder and any client verifies the receipt offline. Reads need no auth. For a single-shot answer to a free-text question, call emem_ask: it routes the question to a place, recalls the relevant bands, and runs the applicable algorithms in one + - [MCP endpoint](https://emem.dev/mcp): JSON-RPC 2.0 over Streamable HTTP; tools/list returns the 18 core tools by default, tier "all" or the /mcp/full endpoint returns all 115. Every tool is callable by name from either endpoint, so narrowing discovery removes no capability; call emem_tools for the map or one tool's schema. Each tool declares one shape in MCP-standard _meta as dev.emem/shape (the form of the answer: scalar, timeseries, raster, geometry, vector, identity, token, proof, plan, fil

https://raw.githubusercontent.com/vortx-ai/emem/HEAD/AGENTS.md

− Rust workspace, 19 crates, version 2.4.2, MSRV 1.91. `emem-guard` is the verdict server for AI inference checkpoints (Anthropic Inference hooks, Claude Code hooks); it is a separate binary and shares no code path with the responder. The bulk of the code− | MCP tool registry (114 tools) | `crates/emem-mcp/src/lib.rs` |+ Rust workspace, 19 crates, version 2.4.3, MSRV 1.91. `emem-guard` is the verdict server for AI inference checkpoints (Anthropic Inference hooks, Claude Code hooks); it is a separate binary and shares no code path with the responder. The bulk of the code+ | MCP tool registry (115 tools) | `crates/emem-mcp/src/lib.rs` |

description

Shared memory for AI agents. One address per fact, one signature you check. No key to read. → Shared, verifiable memory for machines and AI. Encode where data lives; decode with any AI.

version

2.4.2 → 2.4.3