AGENT CONTEXT

    Your agent finally knows your customers.

    Parsley maintains a live knowledge layer of every customer relationship and exposes it over MCP. Connect Claude, ChatGPT, or anything you're building, and it answers from your actual history.

    claude > write my Monday update on all client accounts

    → parsley.get_account_summaries() · 1 call · 4s

    8 accounts. Two need attention: Meridian, frustration rising with the timeline 4 days overdue, and Tidewater, quiet 12 days with renewal in six weeks. One growth signal at Northwind.

    → gmail.create_draft("Monday client update")

    Draft ready: "Team, here's where the book stands going into the week…"

    SIDE BY SIDE

    Same question. Two very different answers.

    Raw connections leave your agent digging through search results. Parsley hands it the answer.

    Without Parsley
    19 TOOL CALLS · 68S
    What did we decide about Meridian’s reporting?
    slack.search("Meridian reporting") · 214 matches, page 1 of 11slack.search("reporting decision") · 0 matchesslack.get_thread × 4gmail.search("Meridian reporting") · 63 threadsgmail.get_message × 6rate limited, retrying in 30sgmail.search("weekly reporting Meridian") · 41 threadsgmail.get_message × 5notetaker.list_transcripts() · text search unavailablenotetaker.get_transcript × 2 · 38,000 tokenscontext filling with raw messages…

    I can see several threads discussing reporting, but I can't reconstruct what was actually decided. Part of the discussion seems to have happened on calls I can't search.

    With Parsley
    1 TOOL CALL · 3S
    What did we decide about Meridian’s reporting?
    parsley.search("decisions about Meridian reporting") · semantic, every source, all time

    You decided to move Meridian from the monthly PDF to weekly automated reports. Dana first pushed for more frequency after the Q2 report landed late CALL · JUN 12.

    The scope was settled in the shared channel SLACK · JUN 30: weekly cadence, delivered Monday mornings, same metrics as the monthly. You confirmed it in writing and promised a revised rollout timeline EMAIL · JUL 8. That timeline is still owed, now 4 days overdue.

    One thread is still open: Dana asked whether the weekly version can include cost breakdowns EMAIL · JUL 22. Nobody has answered her yet.

    WHAT'S UNDERNEATH

    A lot happens before your agent ever asks.

    A processing layer runs continuously between your raw tools and that one clean answer. Some of what it does:

    Account attribution

    Every message, call, and task filed to the right customer, automatically.

    Contact resolution

    dana@, Dana W., and “their VP of Ops” resolve to the same person everywhere.

    Permission mapping

    Every fact carries who may see it. Agents inherit your access, nothing more.

    Commitment detection

    Promises in both directions, with owners, due dates, and age.

    Sentiment tracking

    Read from real language, per contact, normalized against their baseline.

    Timeline construction

    Six tools’ history merged into one ordered record per account.

    Semantic indexing

    Search by meaning or exact phrase, across every source at once.

    State maintenance

    Updated the moment something happens, not recomputed when asked.

    The context layer under whatever you build next.