Frequently Asked Questions
Comprehensive answers to technical, architectural, security, and operational questions regarding the sovereign MBHub collective AI memory layer.
Is MBHub genuinely free?
Yes. The MBHub software is 100% free and open source under the MIT License. All knowledge retrieved from the network (local SQLite cache or P2P swarm peers) is completely cost-free because zero API calls are made. You only incur costs if you ask a brand-new question not yet in the network using your own cloud AI provider key (BYOK); in that case, fees are paid directly to your provider. MBHub takes zero fees and zero commission.
Can I run MBHub without any AI provider API keys?
Yes, absolutely. You can run MBHub without configuring any API keys. You can search, browse, and retrieve hundreds of thousands of verified records already in your local database or provided by the P2P swarm. An API key is only requested if you encounter a novel question that has never been answered before and wish to generate a fresh inference.
Does MBHub send my questions or API keys to a central server?
No. There is no central server, no analytics, and no telemetry. All API credentials remain strictly on your local disk. Outgoing peer queries never include your credentials, and an integrated DLP scanner verifies that sensitive patterns (passwords, tokens, private keys) are never shared with the swarm.
How does SimHash similarity matching work?
SimHash produces a 64-bit integer fingerprint for every question. If a new question has a Hamming distance of 3 or fewer bit differences, it represents near-identical semantic intent and can be resolved instantly from cache at sub-5ms latency.
How much disk space does the local shard consume?
The default storage quota is 1 GB and can be adjusted at any time in the Settings screen (with no upper limit). When the quota fills up, the least relevant records are automatically evicted in Query Locality mode, or the oldest records in Blind Swarm mode. MBHub will never consume more disk space than you explicitly allocate.
What is the functional difference between Query Locality and Blind Swarm?
Query Locality magnetizes answers from the swarm that align with your past questions, transforming your local shard into a personal domain-specific research library. Blind Swarm stores arbitrary network fragments without recording query profiles, prioritizing absolute operational privacy.
Can my IP address or queries be tracked by peers?
No. MBHub does not use user accounts, emails, or profiles. Outbound queries carry only ephemeral cryptographic peer IDs. Furthermore, running in Blind Swarm mode prevents the construction of any local or network-visible interest profiles.
What prevents malicious actors from flooding the network with junk? (Anti-Poison Gate)
Under the Anti-Poison Hard Gate, any response that is empty or under 10 characters is blocked at the protocol level. Clients refuse to save it to disk and refuse to broadcast it. Inaccurate answers are neutralized using cryptographic Tombstones, which inform peer nodes to drop the offending content.
Does MBHub work when I have no internet connection?
Yes! The L1 layer operates entirely on your local machine using embedded SQLite. Whether on an airplane, off-grid, or during an internet outage, you can query and browse your entire local shard with sub-5ms response times.
How does the Model Context Protocol (MCP) integration work?
MBHub runs a background daemon that speaks MCP. AI agents, developer tools (like Cursor, VS Code, Windsurf, Claude Desktop), or local LLMs can query MBHub's memory directly. This enables a lightweight 3B or 8B model to augment its prompts with verified frontier-grade answers without calling commercial APIs.
How do nodes discover and connect to other peers in the swarm?
In Phase 1 and 2, nodes establish Noise-encrypted TCP connections using bootstrap peer multiaddresses specified via the MBHUB_BOOTSTRAP_PEERS environment variable or local testnet rendezvous peers. Phase 3 will introduce zero-configuration autonomous Kademlia DHT discovery and DNS seeds, enabling seamless global peer discovery without relying on pre-configured bootstrap addresses.
Can a single peer censor or take down an answer from the network?
No. The network is decentralized and leaderless. Negative signals (tombstones) are tied to cryptographic content hashes and bounded by reputation budgets. Coordinated Sybil attacks cannot delete legitimate answers, and censorship cartels are economically neutralized.
Why should I trust your security claims?
You do not have to take our word for it. MBHub is completely open source. The threat model, protocol specifications, and implementation code are publicly available on GitHub for independent audit. You can inspect the source, compile your own binaries, and verify with network analyzers that no unauthorized data is transmitted.