FlowSpace

Your stack is a dozen tools pretending to cooperate. CI/CD writes to its database. RevOps writes to theirs. Dashboards query a warehouse fed by brittle ETL. Your wiki holds Tuesday's version of the truth; a chat channel holds Wednesday's; nobody's quite sure which is current. Nothing shares an identity, a schema, or a because-line. Every new integration is a quarter's work. Every audit is a scramble.

GDPR Article 17 right to erasure is not satisfied by a policy document that says "we delete on request." It is satisfied when you can prove the data is gone -- structurally, verifiably, down to the storage address. EU AI Act right-to-explanation is not satisfied by a log file. It is satisfied when the explanation is the data structure itself.

FlowSpace is one platform for all of it. DevOps writes signed records. RevOps writes signed records. Dashboards read signed records. A wiki entry and a chat message land in the same graph as a build log and a sales opportunity — every item content-addressed (BLAKE3 hash — identical content always resolves to the same address, so dedup falls out of the maths), cryptographically signed (you always know who put it there and when), and connected to its sources by because-lines (you always know why it's there and what it draws on). Build logs, product specs, chat transcripts, ledger rows, model weights, rendered PDFs — same primitive, same addressability, same audit.

And the data carries its own intent. Every record knows three things: what it is (signed content), what it relates to (because-line), and where it wants to live (a position the data itself declares — a fixed location, or a function: "with the cohort it serves", "wherever the energy is cheapest right now", "replicated across these regions until 2027"). The platform reads the desire and acts: finds the best-fit host, places it there, re-evaluates as conditions change, migrates if a better fit emerges. No external scheduler. No central placement table. The data is the schedule.

Best-effort placement. FlowSpace doesn't promise perfect placement — it gets as close to the data's stated desire as the cluster allows, and degrades gracefully when constraints tighten. The maths makes near-miss almost-as-good: locality is preserved across whatever dimensions the caller chose to decorate, without anyone having to think about the indexing.

We built this for ourselves first. It runs our own products. We are opening it up because the structural discipline is worth sharing — and because your next integration should be a day's work, not a quarter's.

• Content-addressed storage. BLAKE3 hashes as primary keys. Identical content resolves to the same address, so deduplication falls out of the maths. No duplicate storage costs. Verifiable integrity on every read.
• Ed25519 signatures on everything. Identity is a public key. Signatures are first-class. Every record is authenticated. You always know who put it there and when.
• Because-lines. Every record cites its sources. Walk back from any claim to the evidence that produced it. This is what makes GDPR Article 15 (right of access) and EU AI Act Article 86 (right to explanation) answerable in seconds, not days.
• Typed schemas. Define a data type once; any ingest target matching the schema becomes queryable immediately across all tooling. The next data source is a configuration task, not an engineering project.
• Reusable ingest templates. Common patterns are already wired up (RSS-to-feed, change-feed-to-graph, API-poll-to-schema). Named, signed, revisable.
• Observer-relative trust. Trust is materialised into a time-locked index the observer slides through at query time — no global rank, no stored reputation number.
• Distributed mesh operation. Nodes peer directly. Content follows capacity. No single origin to knock over.
• Privacy-first erasure. Revocation is structural (a rework with valid_until plus key closure). GDPR Article 17 right to erasure is satisfied structurally — provably, down to the storage address. Not by a policy promise you hope someone follows.