Mïmis is an ambitious project to replace the client/server-based mechanism for distributing content that is HTTP with a P2P oriented solution that maintains a massive shared graph.
Initially, Mïmis is focused on making available public resource, and relies on a couple existing projects to manage its state.
The Interplanetary File System is a content-addresses distributed hash table for peer-to-peer file sharing. It is used to store both the data in Mïmis & the update information for synchronizing participants' graphs.
Kubo is Protocol Labs’ Go-based primary implementation of IPFS. It is designed to run locally, and allow users to participate in the network.
To run an IPFS daemon, download Kubo, then run:
ipfs init # if running for the first time
ipfs daemonIn order for Mïmis to access Kubo via the browser, the API endpoint needs to send a CORS Access-Control-Allow-Origin HTTP header. The configuration allows this & to add it using the jq command, do:
KEY=API.HTTPHeaders.Access-Control-Allow-Origin
CURRENT=$(ipfs config $KEY)
VAL=$(jq -c '. + ["http://localhost:5173", "https://mimis.dhappy.org"]' <<< ${CURRENT:-[]})
echo "Setting to '$VAL'"
ipfs config --json $KEY "$VAL"Then restart the IPFS daemon.
Additionally, Kubo allows defining authorzation parameters (both credentials and tokens) and limiting which endpoints those authorizations can access.
Storacha is a CDN for provisioning IPFS resources. To use it, enable it in Mïmis' settings & provide an email address that will be used for authentication.
Veilid is a protocol for anonymous communication with a plan in its roadmap for a DHT. Once that exists, the plan is to also use it for anonymous storage of file information.
Data is stored in a TinkerPop-compatible database which is queried using Gremlin.
IPFS deals specifically with directed acyclic graphs. Preventing cycles in the complex mass of interconnected paths that is Mïmis is essentially impossible, and so the acyclic data structures from IPFS are used to create more complex potentially cyclic ones in the database.
To run Janus Graph for local development, do:
docker compose upTo run with Caddy terminating SSL, as is needed for a secure web socket, run:
docker compose -f docker-compose.yaml -f docker-compose.production.yaml upOnce, JanusGraph is running, you can connect a Gremlin console using:
docker exec -e GREMLIN_REMOTE_HOSTS=janusgraph -it authed-janusgraph bin/gremlin.shWithin the Gremlin console, you connect to the running JanusGraph instance using:
g = traversal().withRemote('conf/remote-graph.properties')
As with most containers, if you just want to poke around the file system:
docker exec -it authed-janusgraph /bin/bashThe data is Mïmis is custom structured to support a variety of different applications.
One of the most fundamental is as a collaborative file system called Argüs allowing users to:
- maintain individual hierarchical collections of information
- create links to incorporate parts of others' collections into their own
- suggest updates to others' resources
- represent resources which served as sources for a work
The most basic idiom in Argus is a directory tree. Nodes in the tree represent conceptual points in the space of all ideas. (In the system, these nodes are “nöopoints” where the sum of all concepts is Vernadsky's “nöosphere”, though they are refered to, for simplicity sake, as “Spots”.) The edges between nodes contains the path elements which are associated with folders in a traditional file system.
Unlike a regular file system where the bulk of the identifying description is in the name of the file, in Argus a Spot may have files of different types associated with it, but the file has no name so to speak. All the identifying information is in the path, & this allows multiple paths to resolve to the same resource without duplicating information. Eventually, ideally, all reasonable paths would resolve correctly.
Path trees, when published, are signed with an Ethereum keypair, and, within the larger system which is Mïmis, they are stored off a Root node by ACCOUNT edges with a signer values. Whenever a new path element or nöopoint is added, it contains the following identifiers:
- the creation timestamp
- the update timestamp
- multihashes of the path name relative to the local root
- the address of the author
Path trees, then, are anchored according to those characteristics. Each specifies a root that already exists in the tree.
In addition to the nöopoints, the path trees also contain mount nodes which connect a path to an existing subtree elsewhere, be it in their own tree or someone else's. The mounts are "union mounts" where several subtrees can be mounted at the same place, and their contents add to each other.
One crucial view of the data is formed by analyzing the graphs as a whole filtered using an Sybil guard like Human Passport. In this larger graph which resource is appropriate for a given path is determined by examining how many verified accounts link to it.
As files are added in locations where there is already an entry, an edge is added from that entry to the previous one, so that the current entry is the one without such a link, but all previous versions can still be accessed.