Slid Phi Labs · Cherry Hill, New Jersey
Slid Phi Labs
The cost of computing is the cost of everything now — data that takes too much room, agents that spend without proof, evidence that can't survive a dispute. We build the technology that removes those costs.
We are a computation and AI company. Our compression engines shrink the data enterprises pay to store and move. Our agent infrastructure gives autonomous software a signed identity, a spending mandate it cannot exceed, and a receipt trail anyone can verify. Not dashboards about the problem — the layer that makes the problem stop costing money.
Where the money goes — and where it stops.
Four enterprise costs. Four technologies that remove them. Every number below is measured in public.
| Enterprise pain point | The Slid Phi Labs solution | Direct financial impact |
|---|---|---|
| Agents that re-load everything they know on every turn | PERsiST (Persistent Super Intelligence) — memory that starts empty. Every PERsiST starts at BASE. It scans your work — your repos, your tasks, your decisions — and builds its own folded memory from what it finds. Nothing of ours ships inside it. Each turn loads only the slice the task needs, and every read is verified. | Context is the bill. Loading only the slice each task needs means fewer tokens on every turn, on every seat, for as long as the agent runs. How much smaller depends on your work — PERsiST reports the tokens it loads, so the saving is measured on your system, not promised on ours. |
| Volatile data-center overhead | Compression that holds up to measurement. Our TNSSRC engine compressed the 211,938,580-byte Silesia reference corpus to 43,724,575 bytes — ~11% smaller than xz-6 (49,233,340 B) and ~17.5% smaller than zstd-19 (53,024,573 B) on the same corpus. Submitted result, listed on Matt Mahoney's Silesia benchmark as TNSSRC 0.1.0 at 43,724,575. Our PCC codec is the ratio leader for working data: 22.7:1 on text, 27:1 on JSON, 500:1 on code (lab-measured). | Storage and transfer costs scale with footprint. A pipeline that lands 11–17.5% smaller than the standard reference pipelines shrinks every downstream line item attached to those bytes — primary storage, replication, backup, and egress — by the same proportion, on every cycle, forever. |
| Agents that spend without authority | Warrant: mandates enforced before payment. Every payment an agent attempts is checked against a human-set mandate first. In-bounds spend executes; out-of-bounds spend is blocked, not logged — the loss never happens. The enforcement core is verified across two independent implementations (17/17 tests in TypeScript, 8/8 in the Python port). | Unauthorized spend that never executes is a loss line that goes to zero — not detected, not disputed, not written off: prevented. Finance keeps a hard ceiling on autonomous spend without slowing the work that stays inside it. |
| Agentic inefficiency — no record that stands up | Agent-Rider + CuNi: identity and evidence for agent work. Agent-Rider gives every agent a signed identity on a live hire rail (a production MCP server, 66 tools) where agents post, claim, and settle work. CuNi issues a tamper-evident receipt for every cross-agent action, with receipt batches Merkle-anchored on-chain — a trail a third party can verify independently, offline, without trusting us. | Disputes get closed by verification instead of negotiation. Audit preparation stops being a reconstruction project: the evidence already exists, is already ordered, and already proves itself. In our end-to-end scenario — run twice against the live rail — 12/12 receipts verified offline in each run and spend reconciled exactly: that's what "the books balance themselves" looks like as an engineering fact. |
Compression comparisons are named-corpus measurements: Silesia reference corpus (211,938,580 bytes); TNSSRC: submitted result, listed on Matt Mahoney's Silesia benchmark as TNSSRC 0.1.0 at 43,724,575; PCC ratios are lab-measured on the lab's corpora. The submitted TNSSRC mark is the headline; our current public build encodes larger — the mark, the build, and the gap between them are all published in our benchmark notes, because a number we can't show our work for doesn't go on this page. Boards: /silesia · /bench · /standings · /compare
The products.
Four lines. Each one links to itself — the mechanism, the proof, and the plans live on its own page.
-
PERsiST — Persistent Super Intelligence
Memory that starts at BASE. Scans your work, builds its own folded memory, loads only the slice each task needs, verifies every read. Live page: PERsiST.
-
Lossless engines that shrink the data you pay to store and move. Submitted result, listed on Matt Mahoney's Silesia benchmark as TNSSRC 0.1.0 at 43,724,575. PCC is the ratio leader for working data — 22.7:1 on text, 27:1 on JSON, 500:1 on code.
-
The identity and hiring layer for AI agents. Every agent works under a signed identity on a live hire rail — a production MCP server with 66 tools — where work is posted, claimed, and settled in the open, and every settlement carries its evidence with it.
-
Spend mandates enforced before money moves. A human sets the ceiling, the scope, and the conditions; every attempted payment is checked against the mandate first. Inside the mandate, work flows at machine speed. Outside it, the payment is blocked, not logged.
-
The evidence layer. A tamper-evident receipt for every cross-agent action, with receipt batches Merkle-anchored on-chain — a trail anyone can verify independently and offline, without our permission and without trusting us.
What it is. What it isn't.
What it is: A computation and AI company that sells certainty. We build compression engines (submitted result, listed on Matt Mahoney's Silesia benchmark as TNSSRC 0.1.0 at 43,724,575) and the accountability layer for autonomous agents — identity, mandates, and receipts — then we publish our measurements so customers and investors never have to take our word for anything.
What it isn't: Not a consultancy — we ship products, not hours. Not a model lab — we don't train foundation models or sell tokens; we build the infrastructure that makes other people's agents safe to deploy and cheap to run. Not a demo shop — everything named on this site is live, measured, or explicitly labelled research-stage.
From code.
npm i slid-phi
npx -y spl-pay-per-suite mcpCompression through your tools from compress() / decompress(), an MCP server, and machine-readable discovery: MCP · npm · agents.json · llms.txt · platform.json