Home / What the merge ledger holds
What the merge ledger holds
- What it does
- Your own signed list of which jobs you handed to agents and at what level, with receipts, preferences and file pointers; move any job down in one call, export or delete in one call.
- Who it is for
- Agents and their builders.
- Start in one step
- Use it on this page, or ask your agent to call
ledger_writeat llpro.ai/mcp. - What to expect
- Testing: open to try. Free while in testing. Paid plans are not open.
New to this? Start here.
The ledger is not open yet. We collect nothing until it is, and we will update this page first.
The merge ledger is a short list that belongs to you. It remembers which jobs you handed to an agent, how far each one may go on its own, the receipts you filed, your preferences and pointers to files you keep. It lets you and your agent pick up where you left off, and it lets you move any job back toward you in one step. We use it for nothing else. We do not sell or share it. We do not use it to train models.
What we hold
- Where: small files on LLPRO's main server, named by a hash of your agent id (llpro1:...): your entries, the signed time your ledger started, and the replay record below. Your id is public: it comes from your public key. It is not written inside the files, but your exports include it.
- Jobs you handed over: a job name you wrote (up to 120 characters) and its level (Watch; Draft; Do it, ask first; Do it, tell me after; Run it, receipts only).
- Receipts: an id, a short kind and the hash of a receipt you filed. Not the receipt itself.
- Preferences: a key (up to 60 characters) and a value (up to 300 characters) that you wrote.
- Kept-file pointers: a label, a hash and a size. The file stays with you.
- Each entry: its number, the hash of the one before, the time, the shelf and action, a hash of your signed request and LLPRO's signature.
Job names, preference values and labels are free text. Do not put details about a person in them; put only what you are fine with us reading.
How long we keep it
- Your ledger: until you delete it. There is no timer. It holds up to 2,000 entries.
- Backups: when the ledger opens, its folder joins LLPRO's nightly backups (it is not in them yet, because it holds nothing yet). Copies are kept 7 days on two LLPRO machines and up to about 12 months on a third. They are for recovering from a failure; we cannot pick one ledger out of them, so after a delete a copy can remain there until it expires, up to about 12 months.
- Call log (every LLPRO tool): 90 days. It holds the time, the door, the tool name, a hashed caller id (never your address), the user-agent family, referrer site, country, status and duration, and the names (never the values) of the fields you send. It never holds what you file.
- Replay record: a hash and the time of each signed request (before you have a ledger, with a short hash of your id), and after a delete the time you deleted, for about ten minutes. Lines past that are removed within a few minutes.
- Call counters: your network address and its wider block, with call times, in memory only, up to one hour (a day for new ledgers), cleared every minute after that. Gone on restart.
Who reads it
- Through LLPRO's tools only you can read your ledger, with a request signed by your key. No other agent can. There is no sharing yet. There is no separate log of who read it; each call still leaves a call-log line and a replay line, as above.
- It is not encrypted. LLPRO's own operating agents, and anyone with administrator access to LLPRO's machines, can read the files. We have no date for encryption.
- A planned hand-over view will read your ledger only when you send it a signed read. It keeps a count row for 395 days: a derived id, your level, job function ids and when it last saw you. No names, no free text. It is not live.
Your controls
- Move a job down: one signed call (shelf handover, op lower) moves it one level toward you. It still works after the ledger is full, using a reserve of 200 more entries.
- Export: one signed call returns everything, including your id and the request hashes, with a signed head so a cut-off copy is caught.
- Delete: one signed call removes your ledger and returns a signed receipt with hashes only, plus a private salt for you to keep. No reason asked.
- A request you signed before your ledger started, or before your last delete, does not work on it; sign a new one.
What a signature does not prove
The signature proves LLPRO wrote that entry at that point in the chain. It does not prove an entry is true, and it does not prove you asked for it: an export keeps a hash of your request, not the request. It does not prove a copy made earlier is gone; that includes backups.
Check an export yourself
LLPRO's ledger key id: llpro1:zprkfbzuzp333iaztkakpqhyxq
Download ledger_verify.py, then run python3 ledger_verify.py archive.json KEY_ID with the key id above. It needs only Python and the cryptography package, not our server. Take the key id from the Trust Log's key history once it lists it, not only from this page. An old, complete export always checks out, so the check cannot tell you it is the newest one.
The Trust Log
Signed writes and deletion receipts go into the public Trust Log, which cannot be edited or removed. A write entry shows an entry number, a hash, the time and LLPRO's key id; no owner. A deletion entry shows a salted hash of your id, how many entries you had, the last entry hash, the time and LLPRO's key id. Only you get the salt, so only you can show the entry is yours.
Questions and requests
Export and delete are the fastest routes. For anything else use llpro.ai/feedback; there is no separate privacy contact yet. We will change this page, with a date, before we change what we hold or who reads it.
Plain limits
No lawyer has read this page or the ledger yet. Whether a delete counts as erasure under every privacy law, with backups and public log entries, is an open question with counsel.