Jevmail for agents and developers

next

Pick the next inbox email. Return the highest scored eligible unhandled INBOX email across connected mailbox copies. Only the latest message in a Gmail thread is eligible: any later message, from the owner or anyone else, makes every earlier message in that thread ineligible, and a thread whose latest message is the owner's is not in the queue. That check is applied when next is read, including after a newer message syncs, and the importance view uses the same eligible set. One score is saved per thread and saved question during sync. Any new message in that thread, received or sent, makes the saved score stale until the thread is rescored; the previous score is not served. Every score uses the same context: that thread's copied headers and snippets, plus the latest message of at most five other open threads in the same mailbox from the last 14 days that share a human From, To, or Cc address. Participants on a message with List-Id or List-Unsubscribe, and no-reply-style addresses, do not count. A new stored message rescores each of those related open threads with that same context, and the previous score is not served once the related context changes. The returned message is the thread's latest eligible message, and its threadId is unchanged. next takes no arguments and never calls Gmail. Before returning a thread, next rescores that thread through the AI Gateway using the same context as every score, including the current time. If it is still the top serving thread, next returns it with the fresh score. If not, next rescores whatever is now top and repeats in that order until a thread survives its rescore. Each new score is saved as that thread's serving score, and each rescore is logged with this call's request id. A model failure returns an error and does not return a stale score. next never sends, never alters drafts, and never changes Gmail. reason.importance_score is that thread score; reason.confidence is null because the individual score has no choice confidence. waiting counts threads in the waiting state; next excludes them until a newer inbound message or undismiss reopens them. unscored counts eligible threads without a fresh serving score; they are retried on later syncs. rescore_in_progress means the saved question changed and the disclosed ranking_question still serves until all replacement scores are ready. The saved question in Settings is the only question. The importance view uses the same persisted order, which stays stable when an email is dismissed. A mailbox may have no selected email; that does not mean it is disconnected. mailboxes and named warnings describe copy coverage, failures and last sync times; incomplete or unscored mail never produces a false empty inbox. After handling a pick, call dismiss with its mailbox and messageId; confirmed Jevmail replies are marked handled automatically. Gmail archiving remains separate; ask before archiving matches from same_sender_in_inbox.

No arguments.

{
  "method": "tools/call",
  "params": {
    "name": "next",
    "arguments": {}
  }
}

echo '{}' | jevmail next --json -

curl -X POST https://jevmail.xyz/api/agent/next \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{}'

{
  "empty": false,
  "mailboxes": [
    {
      "email": "you@example.com",
      "checked": true
    }
  ],
  "message": {
    "mailbox": "you@example.com",
    "messageId": "abc123",
    "subject": "Re: Friday",
    "text": "Friday works?"
  },
  "question": "Which email would Sam be most happy to have responded to?",
  "unscored": 0,
  "rescore_in_progress": false,
  "waiting": 0,
  "reason": {
    "importance_score": 0.8,
    "confidence": null
  },
  "next_step": "Handle this email, call dismiss with its mailbox and messageId, then use same_sender_in_inbox in the same mailbox and ask before archiving other mail; call next again for another pick."
}

dismiss

Mark an inbox thread handled. Immediately mark a copied thread dismissed in Jevmail so next excludes it without waiting for sync. Provide its connected mailbox and exactly one of messageId or threadId. Already handled threads stay unchanged. No Gmail or model call; Gmail labels are unchanged. A newer inbound message reopens the thread. undismiss clears that local dismissal. Logged like every operation.

{
  "method": "tools/call",
  "params": {
    "name": "dismiss",
    "arguments": {
      "mailbox": "you@example.com",
      "messageId": "abc123"
    }
  }
}

echo '{"mailbox":"you@example.com","messageId":"abc123"}' | jevmail dismiss --json -

curl -X POST https://jevmail.xyz/api/agent/dismiss \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"mailbox":"you@example.com","messageId":"abc123"}'

{
  "mailbox": "you@example.com",
  "threadId": "abc456",
  "handled": "dismissed",
  "changed": true
}

undismiss

Clear an inbox thread dismissal. Immediately clear a copied thread's dismissal or wait in Jevmail so its latest message can return in next. Provide its connected mailbox and exactly one of messageId or threadId. A thread that is neither dismissed nor waiting stays unchanged. No Gmail or model call; Gmail labels and message bodies are unchanged. Only the latest message in the thread can be offered, and a thread whose latest message is the owner's stays out of next. Logged like every operation.

{
  "method": "tools/call",
  "params": {
    "name": "undismiss",
    "arguments": {
      "mailbox": "you@example.com",
      "messageId": "abc123"
    }
  }
}

echo '{"mailbox":"you@example.com","messageId":"abc123"}' | jevmail undismiss --json -

curl -X POST https://jevmail.xyz/api/agent/undismiss \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"mailbox":"you@example.com","messageId":"abc123"}'

{
  "mailbox": "you@example.com",
  "threadId": "abc456",
  "handled": "unhandled",
  "changed": true
}

wait

Set an inbox thread to waiting. Mark a copied thread waiting so next skips it without dismissing it. Provide its connected mailbox, a nonempty note of at most 4,000 characters, and exactly one of messageId or threadId. The note is appended to the thread's log with the time and author agent; entries are never edited or deleted. An unhandled thread becomes waiting. A thread already waiting stays waiting, keeps its original waiting time, and gains the new note. A replied or dismissed thread is left unchanged and the note is not added. A newer inbound message reopens the thread, the same way it reopens a dismissal. undismiss also reopens it. No Gmail or model call; Gmail labels and message bodies are unchanged. Logged like every operation.

{
  "method": "tools/call",
  "params": {
    "name": "wait",
    "arguments": {
      "mailbox": "you@example.com",
      "messageId": "abc123",
      "note": "Waiting on competing quotes."
    }
  }
}

echo '{"mailbox":"you@example.com","messageId":"abc123","note":"Waiting on competing quotes."}' | jevmail wait --json -

curl -X POST https://jevmail.xyz/api/agent/wait \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"mailbox":"you@example.com","messageId":"abc123","note":"Waiting on competing quotes."}'

{
  "mailbox": "you@example.com",
  "threadId": "abc456",
  "handled": "waiting",
  "changed": true,
  "notes": [
    {
      "at": 1792000000000,
      "author": "agent",
      "text": "Waiting on competing quotes."
    }
  ]
}

append_note

Append a note to an inbox thread. Append one timestamped note to a copied thread without changing whether it is waiting, dismissed, or unhandled. Provide its connected mailbox, a nonempty note of at most 4,000 characters, and exactly one of messageId or threadId. The response notes are the whole log, oldest first. author is agent for this operation. Entries are never edited or deleted; there is no operation that does either. No Gmail or model call; Gmail labels and message bodies are unchanged. Logged like every operation.

{
  "method": "tools/call",
  "params": {
    "name": "append_note",
    "arguments": {
      "mailbox": "you@example.com",
      "threadId": "abc456",
      "note": "Asked for the second quote."
    }
  }
}

echo '{"mailbox":"you@example.com","threadId":"abc456","note":"Asked for the second quote."}' | jevmail append_note --json -

curl -X POST https://jevmail.xyz/api/agent/append_note \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"mailbox":"you@example.com","threadId":"abc456","note":"Asked for the second quote."}'

{
  "mailbox": "you@example.com",
  "threadId": "abc456",
  "handled": "waiting",
  "notes": [
    {
      "at": 1792000000000,
      "author": "agent",
      "text": "Waiting on competing quotes."
    },
    {
      "at": 1792000001000,
      "author": "agent",
      "text": "Asked for the second quote."
    }
  ]
}

waiting

List waiting inbox threads. List this account's waiting threads, oldest wait first. Each entry has its mailbox, threadId, since (when the wait started, in epoch milliseconds), waitingMs (how long it has been waiting at this read), and notes, oldest first. A note has at, author (agent or owner), and text. Takes no arguments. Never calls Gmail or a model, and does not include message bodies. Logged like every operation.

No arguments.

{
  "method": "tools/call",
  "params": {
    "name": "waiting",
    "arguments": {}
  }
}

echo '{}' | jevmail waiting --json -

curl -X POST https://jevmail.xyz/api/agent/waiting \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{}'

{
  "threads": [
    {
      "mailbox": "you@example.com",
      "threadId": "abc456",
      "since": 1792000000000,
      "waitingMs": 3600000,
      "notes": [
        {
          "at": 1792000000000,
          "author": "agent",
          "text": "Waiting on competing quotes."
        }
      ]
    }
  ]
}

same_sender_in_inbox

Find inbox mail from the same sender or list. Find INBOX mail from the same exact sender address or List-Id in the given connected mailbox’s synced copy, without a Gmail call. Provide mailbox plus either messageId to read a copied source’s From and List-Id headers (the source need not still be in INBOX), or from and/or listId directly. The source is excluded; no other mailbox is searched. Returns mailbox, named warnings, partial, up to 20 newest matches (messageId, from, subject, date), and capped when more matches remain. An incomplete, failed or lapsed copy returns its warning with the last confirmed sync time and partial:true, without falling back to another mailbox. Makes no Gmail writes or Jev calls. Ask the owner before archiving any matches using separately authorized Gmail access.

{
  "method": "tools/call",
  "params": {
    "name": "same_sender_in_inbox",
    "arguments": {
      "mailbox": "you@example.com",
      "from": "sam@example.com"
    }
  }
}

echo '{"mailbox":"you@example.com","from":"sam@example.com"}' | jevmail same_sender_in_inbox --json -

curl -X POST https://jevmail.xyz/api/agent/same_sender_in_inbox \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"mailbox":"you@example.com","from":"sam@example.com"}'

{
  "mailbox": "you@example.com",
  "warnings": [],
  "partial": false,
  "matches": [
    {
      "messageId": "abc123",
      "from": "Sam <sam@example.com>",
      "subject": "Friday plans",
      "date": "2026-10-20T10:00:00.000Z"
    }
  ],
  "capped": false,
  "next_step": "Ask the owner whether to archive these; archive only if they say yes. Jevmail archives none of them."
}

score_draft

Score and store a draft in Jevmail. The default scoring tool. Send up to eight alternatives in one call using shared draft metadata without draft.text and candidates: [{text, subject?}], with messageId or inReplyTo for replies, newEmail: true for new emails, or previousCallId to inherit a stored target. Each is scored sequentially in one shared roundId; new emails follow the existing recipient-and-subject grouping rules; the response contains candidates ranked by overall (newest input on a tie), each with callId, answers and draftReviewUrl, plus bestCallId. A failure stops the batch and reports completed candidates and failedCandidateIndex. Single draft.text requests keep their existing response. Score plain text against the owner's questions and sent-mail evidence, and retain the encrypted draft, answers and revision chain in Jevmail. Never creates, edits or reads a Gmail draft; requires only Gmail read access for evidence. Supply messageId or inReplyTo for a reply, or newEmail: true. When the owner gives feedback that leads to this revision or set of drafts, pass their actual words as ownerFeedback; never invent feedback or infer it from a score. On a revision pass the new draft text and previousCallId=this callId; its mailbox and reply target or newEmail are inherited when omitted. No Gmail tools or requestId needed. Returns per-question native answers; overall, the plain average from 0 to 1 of the ordered answers (a Score level as its share of the last level), with the Choice questions it leaves out listed in overallExcluded; callId, mailbox and workflow. Follow workflow.next_action; a score is not permission to send. Twelve reviews a minute per account, shared with legacy tools. Retrieve by callId with get_scored_draft; save_draft_to_gmail is a separate explicit export.

{
  "method": "tools/call",
  "params": {
    "name": "score_draft",
    "arguments": {
      "messageId": "18c2f0a9b7d3e1f4",
      "draft": {
        "to": [
          "alex@example.com"
        ],
        "subject": "Re: Project timeline",
        "text": "Thanks Alex. Friday works for me."
      }
    }
  }
}

echo '{"messageId":"18c2f0a9b7d3e1f4","draft":{"to":["alex@example.com"],"subject":"Re: Project timeline","text":"Thanks Alex. Friday works for me."}}' | jevmail score_draft --json -

curl -X POST https://jevmail.xyz/api/agent/score_draft \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"messageId":"18c2f0a9b7d3e1f4","draft":{"to":["alex@example.com"],"subject":"Re: Project timeline","text":"Thanks Alex. Friday works for me."}}'

{
  "callId": "8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c",
  "mailbox": "you@example.com",
  "overall": 0.74,
  "overallExcluded": [
    "handling"
  ],
  "questions": [
    {
      "id": "formality",
      "question": "Does the draft's level of formality match how this user writes to these recipients?",
      "type": "noul",
      "noul": 0.81
    }
  ],
  "workflow": {
    "status": "scored",
    "next_action": "present_for_owner_review",
    "review_id": "8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c",
    "revision": 0,
    "progress": "initial"
  }
}

get_scored_draft

Read a stored scored draft. Retrieve the account's immutable Jevmail draft by callId: text, recipients, mailbox, reply target, per-question answers, workflow, previousCallId, ownerFeedback when supplied, revision IDs and review URL. Reads Jevmail only; no Gmail access or writes.

{
  "method": "tools/call",
  "params": {
    "name": "get_scored_draft",
    "arguments": {
      "callId": "8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c"
    }
  }
}

echo '{"callId":"8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c"}' | jevmail get_scored_draft --json -

curl -X POST https://jevmail.xyz/api/agent/get_scored_draft \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"callId":"8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c"}'

{
  "callId": "8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c",
  "draft": {
    "to": [
      "alex@example.com"
    ],
    "cc": [],
    "subject": "Re: Project timeline",
    "text": "Thanks Alex. Friday works for me."
  },
  "mailbox": "you@example.com",
  "previousCallId": null,
  "dismissed": false,
  "sent": null,
  "reviewUrl": "https://jevmail.xyz/drafts/8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c?version=draft",
  "revisions": [
    {
      "callId": "8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c",
      "status": "completed",
      "at": 1792000000000
    }
  ]
}

save_draft_to_gmail

Explicitly save a scored draft to Gmail. An explicit Gmail WRITE. Use only when the owner asks to save to Gmail. Save the Jevmail-held snapshot named by callId. The first save creates one Gmail draft; later revisions in the previousCallId chain update that same Gmail draft. Never create a separate Gmail draft for a revision. Stored scored versions remain unchanged. Never sends or scores. Reuse callId on retries; older versions cannot overwrite a newer export, and uncertain writes never create another copy. External Gmail edits or deletion require owner attention. Returns gmailDraftId, gmailMessageId and mailbox. Requires Gmail compose permission and a text review from score_draft. Gmail edits do not update its scores: paste revised text into score_draft.

{
  "method": "tools/call",
  "params": {
    "name": "save_draft_to_gmail",
    "arguments": {
      "callId": "8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c"
    }
  }
}

echo '{"callId":"8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c"}' | jevmail save_draft_to_gmail --json -

curl -X POST https://jevmail.xyz/api/agent/save_draft_to_gmail \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"callId":"8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c"}'

{
  "callId": "8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c",
  "savedToGmail": true,
  "gmailDraftId": "r-4815162342",
  "gmailMessageId": "18c2f1b0c4e5a6d7",
  "mailbox": "you@example.com"
}

find_conversations

Find conversations. Which conversation a reply to these people would answer, from Gmail metadata only. Without a subject: up to ten of the past year's conversations with every address as a participant, most recently active first, each with the messageId to pass to score_draft, subject, sender, date and message count. With a subject: the conversation to choose explicitly for score_draft (the most recently active one with the same subject once reply prefixes are removed, replying to its latest message from someone else) and up to three alternatives. Thirty lookups a minute per account.

{
  "method": "tools/call",
  "params": {
    "name": "find_conversations",
    "arguments": {
      "to": [
        "alex@example.com"
      ],
      "subject": "Project timeline"
    }
  }
}

echo '{"to":["alex@example.com"],"subject":"Project timeline"}' | jevmail find_conversations --json -

curl -X POST https://jevmail.xyz/api/agent/find_conversations \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"to":["alex@example.com"],"subject":"Project timeline"}'

{
  "match": {
    "messageId": "18c2f0a9b7d3e1f4",
    "threadId": "18c2e7d1a0b9c8f7",
    "subject": "Project timeline",
    "from": "Alex <alex@example.com>",
    "date": "2026-10-12T15:04:00.000Z",
    "messages": 3
  },
  "alternatives": []
}

list_draft_questions

List draft questions. The saved questions every draft is asked, with their IDs, types and criteria. Only the owner can edit them on the Questions page; agents have read-only access to the saved set. There are fourteen by default: twelve Noul questions, one categorical Choice and one Score. At most thirty-two.

No arguments.

{
  "method": "tools/call",
  "params": {
    "name": "list_draft_questions",
    "arguments": {}
  }
}

echo '{}' | jevmail list_draft_questions --json -

curl -X POST https://jevmail.xyz/api/agent/list_draft_questions \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{}'

{
  "questions": [
    {
      "id": "formality",
      "question": "Does the draft's level of formality match how this user writes to these recipients?",
      "type": "noul"
    }
  ]
}

get_draft_test_set

Get the draft test set. The test set for trying question wording: the owner's twenty most recent sent emails, each prepared as if it were a proposed draft with the evidence scoring would retrieve, plus up to ten recent agent drafts from the audit. Returns each case's id, kind (sent or agent_draft), subject, recipients and date. Prepared on first use and kept for six hours; refresh rebuilds it. Pass case ids to try_draft_questions, six at a time.

{
  "method": "tools/call",
  "params": {
    "name": "get_draft_test_set",
    "arguments": {
      "refresh": false
    }
  }
}

echo '{"refresh":false}' | jevmail get_draft_test_set --json -

curl -X POST https://jevmail.xyz/api/agent/get_draft_test_set \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"refresh":false}'

{
  "preparedAt": 1792000000000,
  "cases": [
    {
      "id": "18c2d4e5f6a7b8c9",
      "kind": "sent",
      "subject": "Re: Lunch",
      "to": [
        "sam@example.com"
      ],
      "date": "2026-10-10T12:00:00.000Z"
    }
  ]
}

get_draft_test_case

Get one test case. One test case in full: the draft that is scored and every evidence email the model sees, exactly as prepared. Use it to read why a case scored as it did.

{
  "method": "tools/call",
  "params": {
    "name": "get_draft_test_case",
    "arguments": {
      "caseId": "18c2d4e5f6a7b8c9"
    }
  }
}

echo '{"caseId":"18c2d4e5f6a7b8c9"}' | jevmail get_draft_test_case --json -

curl -X POST https://jevmail.xyz/api/agent/get_draft_test_case \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"caseId":"18c2d4e5f6a7b8c9"}'

{
  "case": {
    "id": "18c2d4e5f6a7b8c9",
    "kind": "sent",
    "subject": "Re: Lunch",
    "state": {
      "world": {
        "thread": [
          "18c2d3a1b2c3d4e5"
        ]
      },
      "emails": {}
    }
  }
}

try_draft_questions

Try draft questions. Score proposed questions on one to six cases of the test set without saving anything. drafts optionally gives a case substitute draft text, scored in place of its own with the same evidence from before the original send; such results say draftOverridden and are never cached. Returns each case's typed answers with distributions and confidence where provided. The legacy scores map contains only yes probabilities; categories and rubric positions have null there. The owner wrote every sent case, so a question about resemblance to their writing should mostly score high on sent cases; a low score is a reason to read that case with get_draft_test_case. Call once per six cases to cover the set. Shares score_draft's limit of twelve a minute.

{
  "method": "tools/call",
  "params": {
    "name": "try_draft_questions",
    "arguments": {
      "questions": [
        "Would I sign off like this?"
      ],
      "caseIds": [
        "18c2d4e5f6a7b8c9"
      ]
    }
  }
}

echo '{"questions":["Would I sign off like this?"],"caseIds":["18c2d4e5f6a7b8c9"]}' | jevmail try_draft_questions --json -

curl -X POST https://jevmail.xyz/api/agent/try_draft_questions \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"questions":["Would I sign off like this?"],"caseIds":["18c2d4e5f6a7b8c9"]}'

{
  "results": [
    {
      "id": "18c2d4e5f6a7b8c9",
      "scores": {
        "trial_1": 0.88
      },
      "answers": {
        "trial_1": {
          "type": "noul",
          "noul": 0.88
        }
      }
    }
  ]
}

list_calls

List this account's jevmail calls. The account's call log, newest first: every call the owner or their agents made to jevmail, over MCP, the agent key or the site (not requests the site makes on its own to draw a page). Each line has id, at, call, tool, status, httpStatus, durationMs, gmailRequests (Gmail API requests jevmail made answering it) and draftCallId for a scoring call. limit is 1–100 (default 20); pass nextBefore as before for the next page. Read one with get_call. Reads Jevmail only.

{
  "method": "tools/call",
  "params": {
    "name": "list_calls",
    "arguments": {
      "limit": 1
    }
  }
}

echo '{"limit":1}' | jevmail list_calls --json -

curl -X POST https://jevmail.xyz/api/agent/list_calls \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"limit":1}'

{
  "calls": [
    {
      "id": "c0ffee00-1111-4222-8333-444455556666",
      "at": "2026-10-13T20:31:00.000Z",
      "call": "agent_draft_assess",
      "tool": "score_draft",
      "status": "ok",
      "httpStatus": 200,
      "durationMs": 5400,
      "gmailRequests": 6,
      "draftCallId": "8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c"
    }
  ],
  "nextBefore": "c0ffee00-1111-4222-8333-444455556666",
  "logUrl": "https://jevmail.xyz/log"
}

get_call

Read one jevmail call. One call from list_calls by id: what it was asked (input), what jevmail answered (output), each side as JSON when it was whole JSON and marked when cut at 64 KB, the Gmail API requests jevmail made answering it (method, result, time, pacing wait, quota units; never IDs or content), and models: each Jev model request exactly as sent and Jev's response exactly as received (kept whole up to 256 KB a side). Read top to bottom: input, gmail, models, output. Kept 90 days. Reads Jevmail only.

{
  "method": "tools/call",
  "params": {
    "name": "get_call",
    "arguments": {
      "id": "c0ffee00-1111-4222-8333-444455556666"
    }
  }
}

echo '{"id":"c0ffee00-1111-4222-8333-444455556666"}' | jevmail get_call --json -

curl -X POST https://jevmail.xyz/api/agent/get_call \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"id":"c0ffee00-1111-4222-8333-444455556666"}'

{
  "id": "c0ffee00-1111-4222-8333-444455556666",
  "tool": "score_draft",
  "status": "ok",
  "input": {
    "newEmail": true
  },
  "output": {
    "callId": "8f0c2a7e-3b1d-4c5e-9f6a-1d2e3f4a5b6c"
  },
  "gmail": [
    {
      "method": "messages.get",
      "outcome": "success",
      "status": 200,
      "durationMs": 80
    }
  ],
  "models": [],
  "url": "https://jevmail.xyz/log/c0ffee00-1111-4222-8333-444455556666"
}

get_profile

Whose account this is. The account the token speaks for: a stable id, the Gmail address and the owner's name.

No arguments.

{
  "method": "tools/call",
  "params": {
    "name": "get_profile",
    "arguments": {}
  }
}

echo '{}' | jevmail get_profile --json -

curl -X POST https://jevmail.xyz/api/agent/get_profile \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{}'

{
  "id": "u_7f3a",
  "email": "you@example.com",
  "name": "Your Name"
}

call_log

Read operation request ids. Read this account's operation calls. Pass requestId alone for one call, or omit it to list recent calls. A list can filter by operation, threadId, messageId, and an inclusive UTC since and until; limit is 1–100 (default 50) and the rows come back oldest first. Each row has the request id, time, operation, caller, mailbox, the message and thread ids the call was given, the ids it returned, and ok or an error code. A next call read by request id also includes rescores, oldest first: the request id, time, mailbox, thread id and message id of each thread next rescored. Rows contain no subjects, bodies or sender addresses, and only the account that made the call can read them. Reads Jevmail only.

{
  "method": "tools/call",
  "params": {
    "name": "call_log",
    "arguments": {
      "requestId": "req_BBBBBBBBBBBBBBBBBBBBBB"
    }
  }
}

echo '{"requestId":"req_BBBBBBBBBBBBBBBBBBBBBB"}' | jevmail call_log --json -

curl -X POST https://jevmail.xyz/api/agent/call_log \
  -H "Authorization: Bearer $JEVMAIL_AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"requestId":"req_BBBBBBBBBBBBBBBBBBBBBB"}'

{
  "requestId": "req_AAAAAAAAAAAAAAAAAAAAAA",
  "call": {
    "requestId": "req_BBBBBBBBBBBBBBBBBBBBBB",
    "at": "2026-10-27T19:00:00.000Z",
    "operation": "next",
    "caller": "owner",
    "mailbox": "you@example.com",
    "inputMessageIds": [],
    "inputThreadIds": [],
    "outcome": "ok",
    "outputMessageIds": [
      "abc123"
    ],
    "outputThreadIds": [
      "abc456"
    ]
  },
  "rescores": [
    {
      "requestId": "req_BBBBBBBBBBBBBBBBBBBBBB",
      "at": "2026-10-27T19:00:00.000Z",
      "mailbox": "you@example.com",
      "threadId": "abc456",
      "messageId": "abc123"
    }
  ]
}