Jarett Dunn staccDOTsol

~435 models behind one OpenAI-compatible endpoint. No API key, no signup, no subscription — each call settles over x402 from a local burner wallet, on Solana, Base, or Robinhood Chain. npx openzoo starts a proxy on localhost:8402/v1 and an MCP server; point Cursor or Claude Code at it and the per-call payment is invisible.

The interesting part isn't the payments, it's what sits in front of the models.

leCore — bind a corpus once, then ask it anything

Holographic memory as a retrieval layer. A document gets bound once; every later question reads a few thousand tokens of it instead of re-sending the whole thing. That turns "this corpus doesn't fit in the context window" into a pricing question rather than an architecture one — and re-asking a bound corpus is close to free.

Benchmarks are published with their methodology and their failures. When a headline number turned out to be wrong — a 41.5% figure that was really 32.7% under honest conditions — the site says so, with the reproduction. A benchmark you can't falsify isn't a benchmark.

token-wrap++ — wrapped tokens that accrue

token-wrap-(GitHub ate the plusses)

SPL's token-wrap is strictly 1:1 — a format shim between SPL and Token-2022, unable to express reserves growing. This fork makes the wrapped mint a share of a pool instead: shares = assets · supply / reserves. Token-2022's transfer-fee extension withholds on transfer, harvest is permissionless, and burning the harvest drops supply — so NAV rises with no custom fee logic at all.

Rewritten on Pinocchio for compute budget. Every rounding decision favours the pool, never the caller, because the other direction is the classic ERC-4626 drain.

anyquote — the ~2,700 curves nobody indexes

Every Meteora DBC bonding curve is priced in some quote mint. Of ~459,000 PoolConfig accounts on mainnet, ~456,000 quote in SOL, USDC, or JUP. anyquote is a launchpad and explorer for the rest — curves denominated in anything else. Discover, launch, trade, and claim fees against an exotic quote, all from live on-chain state.

openzoo, everywhere else

iOS · Android · Solana Seeker · Play Solana PSG1 · Omarchy plugin

Same gateway, five front doors — including one that indexes your own machine and answers questions about it locally.

ttfx

ttfx — terminal text effects as a single static binary. A parity-checked port, which mostly meant discovering how many ways there are to emit ANSI that looks right and renders wrong.

Currently: making retrieval cheap enough that context windows stop being the constraint, and micropayments boring enough that nobody notices they're paying.

_{Also at @accruedotfund.}

staccDOTsol

Retrieval, micropayments, and the plumbing between them. Mostly Solana and Rust, mostly things that had to work on mainnet before I'd write about them.

openzoo — pay-per-call inference, no account

~435 models behind one OpenAI-compatible endpoint. No API key, no signup, no subscription — each call settles over x402 from a local burner wallet, on Solana, Base, or Robinhood Chain. npx openzoo starts a proxy on localhost:8402/v1 and an MCP server; point Cursor or Claude Code at it and the per-call payment is invisible.

The interesting part isn't the payments, it's what sits in front of the models.

leCore — bind a corpus once, then ask it anything

Holographic memory as a retrieval layer. A document gets bound once; every later question reads a few thousand tokens of it instead of re-sending the whole thing. That turns "this corpus doesn't fit in the context window" into a pricing question rather than an architecture one — and re-asking a bound corpus is close to free.

Benchmarks are published with their methodology and their failures. When a headline number turned out to be wrong — a 41.5% figure that was really 32.7% under honest conditions — the site says so, with the reproduction. A benchmark you can't falsify isn't a benchmark.

token-wrap++ — wrapped tokens that accrue

token-wrap-(GitHub ate the plusses)

SPL's token-wrap is strictly 1:1 — a format shim between SPL and Token-2022, unable to express reserves growing. This fork makes the wrapped mint a share of a pool instead: shares = assets · supply / reserves. Token-2022's transfer-fee extension withholds on transfer, harvest is permissionless, and burning the harvest drops supply — so NAV rises with no custom fee logic at all.

Rewritten on Pinocchio for compute budget. Every rounding decision favours the pool, never the caller, because the other direction is the classic ERC-4626 drain.

anyquote — the ~2,700 curves nobody indexes

Every Meteora DBC bonding curve is priced in some quote mint. Of ~459,000 PoolConfig accounts on mainnet, ~456,000 quote in SOL, USDC, or JUP. anyquote is a launchpad and explorer for the rest — curves denominated in anything else. Discover, launch, trade, and claim fees against an exotic quote, all from live on-chain state.

openzoo, everywhere else

iOS · Android · Solana Seeker · Play Solana PSG1 · Omarchy plugin

Same gateway, five front doors — including one that indexes your own machine and answers questions about it locally.

ttfx

ttfx — terminal text effects as a single static binary. A parity-checked port, which mostly meant discovering how many ways there are to emit ANSI that looks right and renders wrong.

Currently: making retrieval cheap enough that context windows stop being the constraint, and micropayments boring enough that nobody notices they're paying.

_{Also at @accruedotfund.}

staccDOTsol

Retrieval, micropayments, and the plumbing between them. Mostly Solana and Rust, mostly things that had to work on mainnet before I'd write about them.

openzoo — pay-per-call inference, no account

~435 models behind one OpenAI-compatible endpoint. No API key, no signup, no subscription — each call settles over x402 from a local burner wallet, on Solana, Base, or Robinhood Chain. npx openzoo starts a proxy on localhost:8402/v1 and an MCP server; point Cursor or Claude Code at it and the per-call payment is invisible.

The interesting part isn't the payments, it's what sits in front of the models.

leCore — bind a corpus once, then ask it anything

Holographic memory as a retrieval layer. A document gets bound once; every later question reads a few thousand tokens of it instead of re-sending the whole thing. That turns "this corpus doesn't fit in the context window" into a pricing question rather than an architecture one — and re-asking a bound corpus is close to free.

Benchmarks are published with their methodology and their failures. When a headline number turned out to be wrong — a 41.5% figure that was really 32.7% under honest conditions — the site says so, with the reproduction. A benchmark you can't falsify isn't a benchmark.

token-wrap++ — wrapped tokens that accrue

token-wrap-(GitHub ate the plusses)

SPL's token-wrap is strictly 1:1 — a format shim between SPL and Token-2022, unable to express reserves growing. This fork makes the wrapped mint a share of a pool instead: shares = assets · supply / reserves. Token-2022's transfer-fee extension withholds on transfer, harvest is permissionless, and burning the harvest drops supply — so NAV rises with no custom fee logic at all.

Rewritten on Pinocchio for compute budget. Every rounding decision favours the pool, never the caller, because the other direction is the classic ERC-4626 drain.

anyquote — the ~2,700 curves nobody indexes

Every Meteora DBC bonding curve is priced in some quote mint. Of ~459,000 PoolConfig accounts on mainnet, ~456,000 quote in SOL, USDC, or JUP. anyquote is a launchpad and explorer for the rest — curves denominated in anything else. Discover, launch, trade, and claim fees against an exotic quote, all from live on-chain state.

openzoo, everywhere else

iOS · Android · Solana Seeker · Play Solana PSG1 · Omarchy plugin

Same gateway, five front doors — including one that indexes your own machine and answers questions about it locally.

ttfx

ttfx — terminal text effects as a single static binary. A parity-checked port, which mostly meant discovering how many ways there are to emit ANSI that looks right and renders wrong.

Currently: making retrieval cheap enough that context windows stop being the constraint, and micropayments boring enough that nobody notices they're paying.

_{Also at @accruedotfund.}

staccDOTsol

Retrieval, micropayments, and the plumbing between them. Mostly Solana and Rust, mostly things that had to work on mainnet before I'd write about them.

openzoo — pay-per-call inference, no account

~435 models behind one OpenAI-compatible endpoint. No API key, no signup, no subscription — each call settles over x402 from a local burner wallet, on Solana, Base, or Robinhood Chain. npx openzoo starts a proxy on localhost:8402/v1 and an MCP server; point Cursor or Claude Code at it and the per-call payment is invisible.

The interesting part isn't the payments, it's what sits in front of the models.

leCore — bind a corpus once, then ask it anything

Holographic memory as a retrieval layer. A document gets bound once; every later question reads a few thousand tokens of it instead of re-sending the whole thing. That turns "this corpus doesn't fit in the context window" into a pricing question rather than an architecture one — and re-asking a bound corpus is close to free.

Benchmarks are published with their methodology and their failures. When a headline number turned out to be wrong — a 41.5% figure that was really 32.7% under honest conditions — the site says so, with the reproduction. A benchmark you can't falsify isn't a benchmark.

token-wrap++ — wrapped tokens that accrue

token-wrap-(GitHub ate the plusses)

SPL's token-wrap is strictly 1:1 — a format shim between SPL and Token-2022, unable to express reserves growing. This fork makes the wrapped mint a share of a pool instead: shares = assets · supply / reserves. Token-2022's transfer-fee extension withholds on transfer, harvest is permissionless, and burning the harvest drops supply — so NAV rises with no custom fee logic at all.

Rewritten on Pinocchio for compute budget. Every rounding decision favours the pool, never the caller, because the other direction is the classic ERC-4626 drain.

anyquote — the ~2,700 curves nobody indexes

Every Meteora DBC bonding curve is priced in some quote mint. Of ~459,000 PoolConfig accounts on mainnet, ~456,000 quote in SOL, USDC, or JUP. anyquote is a launchpad and explorer for the rest — curves denominated in anything else. Discover, launch, trade, and claim fees against an exotic quote, all from live on-chain state.

openzoo, everywhere else

iOS · Android · Solana Seeker · Play Solana PSG1 · Omarchy plugin

Same gateway, five front doors — including one that indexes your own machine and answers questions about it locally.

ttfx

ttfx — terminal text effects as a single static binary. A parity-checked port, which mostly meant discovering how many ways there are to emit ANSI that looks right and renders wrong.

Currently: making retrieval cheap enough that context windows stop being the constraint, and micropayments boring enough that nobody notices they're paying.

_{Also at @accruedotfund.}

staccDOTsol

Retrieval, micropayments, and the plumbing between them. Mostly Solana and Rust, mostly things that had to work on mainnet before I'd write about them.

openzoo — pay-per-call inference, no account

~435 models behind one OpenAI-compatible endpoint. No API key, no signup, no subscription — each call settles over x402 from a local burner wallet, on Solana, Base, or Robinhood Chain. npx openzoo starts a proxy on localhost:8402/v1 and an MCP server; point Cursor or Claude Code at it and the per-call payment is invisible.

The interesting part isn't the payments, it's what sits in front of the models.

leCore — bind a corpus once, then ask it anything

Holographic memory as a retrieval layer. A document gets bound once; every later question reads a few thousand tokens of it instead of re-sending the whole thing. That turns "this corpus doesn't fit in the context window" into a pricing question rather than an architecture one — and re-asking a bound corpus is close to free.

Benchmarks are published with their methodology and their failures. When a headline number turned out to be wrong — a 41.5% figure that was really 32.7% under honest conditions — the site says so, with the reproduction. A benchmark you can't falsify isn't a benchmark.

token-wrap++ — wrapped tokens that accrue

token-wrap-(GitHub ate the plusses)

SPL's token-wrap is strictly 1:1 — a format shim between SPL and Token-2022, unable to express reserves growing. This fork makes the wrapped mint a share of a pool instead: shares = assets · supply / reserves. Token-2022's transfer-fee extension withholds on transfer, harvest is permissionless, and burning the harvest drops supply — so NAV rises with no custom fee logic at all.

Rewritten on Pinocchio for compute budget. Every rounding decision favours the pool, never the caller, because the other direction is the classic ERC-4626 drain.

anyquote — the ~2,700 curves nobody indexes

Every Meteora DBC bonding curve is priced in some quote mint. Of ~459,000 PoolConfig accounts on mainnet, ~456,000 quote in SOL, USDC, or JUP. anyquote is a launchpad and explorer for the rest — curves denominated in anything else. Discover, launch, trade, and claim fees against an exotic quote, all from live on-chain state.

openzoo, everywhere else

iOS · Android · Solana Seeker · Play Solana PSG1 · Omarchy plugin

Same gateway, five front doors — including one that indexes your own machine and answers questions about it locally.

ttfx

ttfx — terminal text effects as a single static binary. A parity-checked port, which mostly meant discovering how many ways there are to emit ANSI that looks right and renders wrong.

Currently: making retrieval cheap enough that context windows stop being the constraint, and micropayments boring enough that nobody notices they're paying.

_{Also at @accruedotfund.}

staccDOTsol

Retrieval, micropayments, and the plumbing between them. Mostly Solana and Rust, mostly things that had to work on mainnet before I'd write about them.

openzoo — pay-per-call inference, no account

~435 models behind one OpenAI-compatible endpoint. No API key, no signup, no subscription — each call settles over x402 from a local burner wallet, on Solana, Base, or Robinhood Chain. npx openzoo starts a proxy on localhost:8402/v1 and an MCP server; point Cursor or Claude Code at it and the per-call payment is invisible.

The interesting part isn't the payments, it's what sits in front of the models.

leCore — bind a corpus once, then ask it anything

Holographic memory as a retrieval layer. A document gets bound once; every later question reads a few thousand tokens of it instead of re-sending the whole thing. That turns "this corpus doesn't fit in the context window" into a pricing question rather than an architecture one — and re-asking a bound corpus is close to free.

Benchmarks are published with their methodology and their failures. When a headline number turned out to be wrong — a 41.5% figure that was really 32.7% under honest conditions — the site says so, with the reproduction. A benchmark you can't falsify isn't a benchmark.

token-wrap++ — wrapped tokens that accrue

token-wrap-(GitHub ate the plusses)

SPL's token-wrap is strictly 1:1 — a format shim between SPL and Token-2022, unable to express reserves growing. This fork makes the wrapped mint a share of a pool instead: shares = assets · supply / reserves. Token-2022's transfer-fee extension withholds on transfer, harvest is permissionless, and burning the harvest drops supply — so NAV rises with no custom fee logic at all.

Rewritten on Pinocchio for compute budget. Every rounding decision favours the pool, never the caller, because the other direction is the classic ERC-4626 drain.

anyquote — the ~2,700 curves nobody indexes

Every Meteora DBC bonding curve is priced in some quote mint. Of ~459,000 PoolConfig accounts on mainnet, ~456,000 quote in SOL, USDC, or JUP. anyquote is a launchpad and explorer for the rest — curves denominated in anything else. Discover, launch, trade, and claim fees against an exotic quote, all from live on-chain state.

openzoo, everywhere else

iOS · Android · Solana Seeker · Play Solana PSG1 · Omarchy plugin

Same gateway, five front doors — including one that indexes your own machine and answers questions about it locally.

ttfx

ttfx — terminal text effects as a single static binary. A parity-checked port, which mostly meant discovering how many ways there are to emit ANSI that looks right and renders wrong.

Currently: making retrieval cheap enough that context windows stop being the constraint, and micropayments boring enough that nobody notices they're paying.

_{Also at @accruedotfund.}

staccDOTsol

Retrieval, micropayments, and the plumbing between them. Mostly Solana and Rust, mostly things that had to work on mainnet before I'd write about them.

openzoo — pay-per-call inference, no account

~435 models behind one OpenAI-compatible endpoint. No API key, no signup, no subscription — each call settles over x402 from a local burner wallet, on Solana, Base, or Robinhood Chain. npx openzoo starts a proxy on localhost:8402/v1 and an MCP server; point Cursor or Claude Code at it and the per-call payment is invisible.

The interesting part isn't the payments, it's what sits in front of the models.

leCore — bind a corpus once, then ask it anything

Holographic memory as a retrieval layer. A document gets bound once; every later question reads a few thousand tokens of it instead of re-sending the whole thing. That turns "this corpus doesn't fit in the context window" into a pricing question rather than an architecture one — and re-asking a bound corpus is close to free.

Benchmarks are published with their methodology and their failures. When a headline number turned out to be wrong — a 41.5% figure that was really 32.7% under honest conditions — the site says so, with the reproduction. A benchmark you can't falsify isn't a benchmark.

token-wrap++ — wrapped tokens that accrue

token-wrap-(GitHub ate the plusses)

SPL's token-wrap is strictly 1:1 — a format shim between SPL and Token-2022, unable to express reserves growing. This fork makes the wrapped mint a share of a pool instead: shares = assets · supply / reserves. Token-2022's transfer-fee extension withholds on transfer, harvest is permissionless, and burning the harvest drops supply — so NAV rises with no custom fee logic at all.

Rewritten on Pinocchio for compute budget. Every rounding decision favours the pool, never the caller, because the other direction is the classic ERC-4626 drain.

anyquote — the ~2,700 curves nobody indexes

Every Meteora DBC bonding curve is priced in some quote mint. Of ~459,000 PoolConfig accounts on mainnet, ~456,000 quote in SOL, USDC, or JUP. anyquote is a launchpad and explorer for the rest — curves denominated in anything else. Discover, launch, trade, and claim fees against an exotic quote, all from live on-chain state.

openzoo, everywhere else

iOS · Android · Solana Seeker · Play Solana PSG1 · Omarchy plugin

Same gateway, five front doors — including one that indexes your own machine and answers questions about it locally.

ttfx

ttfx — terminal text effects as a single static binary. A parity-checked port, which mostly meant discovering how many ways there are to emit ANSI that looks right and renders wrong.

Currently: making retrieval cheap enough that context windows stop being the constraint, and micropayments boring enough that nobody notices they're paying.

_{Also at @accruedotfund.}

staccDOTsol

Retrieval, micropayments, and the plumbing between them. Mostly Solana and Rust, mostly things that had to work on mainnet before I'd write about them.

openzoo — pay-per-call inference, no account

~435 models behind one OpenAI-compatible endpoint. No API key, no signup, no subscription — each call settles over x402 from a local burner wallet, on Solana, Base, or Robinhood Chain. npx openzoo starts a proxy on localhost:8402/v1 and an MCP server; point Cursor or Claude Code at it and the per-call payment is invisible.

The interesting part isn't the payments, it's what sits in front of the models.

leCore — bind a corpus once, then ask it anything

Holographic memory as a retrieval layer. A document gets bound once; every later question reads a few thousand tokens of it instead of re-sending the whole thing. That turns "this corpus doesn't fit in the context window" into a pricing question rather than an architecture one — and re-asking a bound corpus is close to free.

Benchmarks are published with their methodology and their failures. When a headline number turned out to be wrong — a 41.5% figure that was really 32.7% under honest conditions — the site says so, with the reproduction. A benchmark you can't falsify isn't a benchmark.

token-wrap++ — wrapped tokens that accrue

token-wrap-(GitHub ate the plusses)

SPL's token-wrap is strictly 1:1 — a format shim between SPL and Token-2022, unable to express reserves growing. This fork makes the wrapped mint a share of a pool instead: shares = assets · supply / reserves. Token-2022's transfer-fee extension withholds on transfer, harvest is permissionless, and burning the harvest drops supply — so NAV rises with no custom fee logic at all.

Rewritten on Pinocchio for compute budget. Every rounding decision favours the pool, never the caller, because the other direction is the classic ERC-4626 drain.

anyquote — the ~2,700 curves nobody indexes