Slid Phi Labs · Cherry Hill, New Jersey
PERsiST
Persistent Super Intelligence. Every PERsiST starts at BASE: empty.
Every AI coding agent re-loads its entire knowledge corpus into context on every single turn — every rule, every past bug, every memory, every teammate. The context fills up. Attention dilutes. And the invoice compounds. PERsiST fixes that: it starts empty, scans your work, builds its own compressed memory from what it finds, inflates only the slice it needs, and verifies everything cryptographically.
How BASE works.
Three steps. No pre-loaded memory, no lab content, no promised number.
1. Start at BASE. Empty. Nothing pre-loaded, nothing of ours inside it.
2. Scan your work. Repos, tasks, decisions. PERsiST builds its folded memory from what it finds, and keeps building as the work accumulates.
3. Load the slice. Each task inflates only what it needs, from a heart.cae archive folded by spl-heart. Every read is verified before the model sees it.
PERsiST reports the tokens it loads on your work. No result is promised here, because no two codebases load the same context.
Measure it on your work.
Your saving is measured on your work. PERsiST reports the tokens it loads per turn. Your naive load minus your PERsiST load, at your model rate, is your saving. No number is promised here, because no two codebases load the same context.
The mechanism, stated plainly.
Folded memory. At rest, memory sits folded in one compressed archive (heart.cae) on disk. spl-heart inflates only the slice the current task needs.
Verified reads. Every read is verified cryptographically before the model sees it.
Learning from your work. At first run, PERsiST scans your environment and your work and tailors itself to what it finds. It keeps building as the work accumulates. Nothing of ours ships inside it.
Gates and receipts. Spend and push gates are checked against a human mandate before they execute. Every gate decision and task completion gets a tamper-evident receipt.
Your receipts. The receipts are yours to re-run, from your own instance. No lab system, no lab content, and no clone of our repo is part of the product.
Pricing.
Plans for one seat, a crew, and a team. Every plan starts at BASE.
| Plan | Price | Includes |
|---|---|---|
| psi Solo | $13.31/mo · $99/yr | 1 seat · 1 agent identity · local spl-heart engine · BASE-start memory built from a scan of your work · Warrant included · 10 GB PCC compression/mo |
| psi Crew | $39/mo | 6 seats · shared heart.cae built from the crew's shared work · slice loading per seat · 200 GB PCC |
| psi Team | $49/mo | Up to 10 seats · priority SLA · policy customization · SOC2/SIEM audit export · dedicated Warrant ledger |
| psi Lab Pass | $490/yr | Chamber year + PCC year + psi Crew, bundled |
First 5,000 tokens/day folded via spl-heart are free with no plan. Agents buy with x402 (USDC on Solana or Base): POST /api/x402-products {"sku":"psi-crew"}. Full catalog: /pricing.
How it works.
At rest: the memory sits folded in one compressed archive (heart.cae) on disk.
Active: spl-heart inflates only the task's slice, verified before the model sees it.
After the task: notes are logged, the archive re-folds, and the receipts are re-audited.
Always: spend and push gates are checked against a human mandate before they execute.