Solana mainnet,
paused on a frame.

Three jobs that used to cost a day each and a fragile setup.

Replay collapses every one to the same loop: boot, drive, read. On a throwaway fork of real mainnet you can run ten times before lunch — no prod risk, no flaky local repro, no waiting on anyone.

One click on a slot, one click on Start. You are now Solana mainnet at that exact frame, with no other tenants.

Cached snapshots skip the network fetch, then still extract and index locally. Uncached ones fetch first. Pick the compatibility version. Pick a Geyser plugin (or none). The Replay environment that comes up is isolated to you, in the region you chose.

USDC. Jupiter. Your wallet. The Raydium AMM that broke at 9pm. Open the address, see the bank's actual answer.

The sidebar is your production address book — quick-add chips for the programs you'll always need, paste-anywhere input for the ones you won't. The detail panel reads the bank: owner, balance, data size, rent epoch, raw bytes. Mutate lamports, owner, data, or executable in-place — every patch is labeled, and the moment one lands the environment flips to a labeled local fork. Re-boot to return to canonical mainnet.

Append at block_end. Inject after a known signature. Replace the transaction at slot index 14.

Eight anchors. One queue. Run advance and canonical mainnet blocks stream in, with your transactions woven in at the slot you asked for. Replay walks forward block-by-block — canonical, then yours, then canonical again — exactly the order it applies them. Cancel any time and inspect what landed.

Walk the chain forward from your boot point. 1,446 txs in a block, your splice at index 10, the fee at 1,005,000 lamports.

Filter by signature or any pubkey in the tx — recipient, signer, program. The list reflects your replay chain, not upstream mainnet, so the tx you spliced in at slot 418,908,420 sits exactly where you put it. Click any row for the decoded instructions, log messages, and pre/post balances.

Click any committed tx and the State diff tab shows the exact pre/post account state the validator wrote — lamports, owner, data — with a hex view that highlights the bytes that changed.

Every non-vote transaction your replay session executes is captured as a state-diff fixture: writable rows side-by-side with role-only rows, lamport movements stacked, owner changes inline, and a hex diff for any account whose data moved. Identical wire shape on the API (GET /fixtures/tx/:sig), the MCP tool (replay_fixture_get), and the k256-replay CLI (k256-replay diff <sig>) — drop the same signature into any one and you'll see the same diff. Iterate fearlessly: clear the cache, advance again, diff again. A 5 GiB FIFO bound and a process-lifetime evicted/oversized/cleared counter keep you honest about what's still in the session.

Checkpoint before a risky splice. Try three approaches. Return to the exact state you started from — as many times as you want.

Every checkpoint is a complete, independent save of your environment at that slot. Restoring is instant. Keep as many as you need — name them, restore any one, delete the ones you're done with. Experiment without fear: there's always a way back to a known-good state.

Every read endpoint your existing tools already speak — hidden until you reveal and copy it.

@solana/web3.js, anchor, solana-cli, your indexer, your bots, your dashboards. Reveal the direct endpoint when you need it, copy the curl, and treat that URL like an operational secret. Subscribe to slotSubscribe, then advance the fork to produce the next local slot. Yellowstone gRPC is one switch away on boot. Nothing to rewrite.

Start slot, target slot, duration, status. Twenty most recent jobs, in memory.

Skim what ran, expand a row for the exit code, the finality choice, and the full job ID. The signal you want when you're iterating on a splice plan and need to know which run produced which state.