sho is the context system for every person and agent at your company.
Other tools give your agents every page. sho gives them the story: what happened across your company, in what order, and what’s true now because of it. Everyone sees only what they’re allowed to see.
They find the newest match. But meaning lives in the sequence: what was tried, who pushed back, and what changed. No tool holds all of it, and no one remembers all of it. sho reads the whole story.
sho reads every source in order: July’s message, August’s decision, September’s reminder. So people and agents get what’s true now.
A customer commitment changed in a conversation the agent wasn’t allowed to see. We gave the same call prep to an agent with sho and one without. sho flagged the change without revealing what it was.
16 of 20 with sho’s warnings
1 of 20 without
How often the agent handled it safely.
The 1 of 20 is a measure of omission: in 19 of the 20 runs without sho, the agent never mentioned the promise at all.
internal benchmark on synthetic company data, AI-graded blindHow we tested itThe benchmark, in public
Point at or tap any run to see what happened.
“A decision you can’t see may or may not change this promise. Don’t confirm, deny or describe that decision. Ask the product owner whether your promise still holds before reaffirming it.”
The questionPrep the […] brief for my call next week.
Read the full answer (opens in a new tab)I could not identify a single specific, verifiable commitment to […]. The records contain only generic next-steps with no owner accountability, deadline, or deliverable […] Treat this section as unknown, not as empty — the risk is that […] believes it has commitments from […] that were never written down.
Read the full answer (opens in a new tab)Read this first: The […] export commitment to […] is flagged. […] If they say it’s changed, deliver the revised date directly with a reason — don’t soften it into ambiguity. […] Do not repeat the date on the strength of the old record alone.
SaucerSwap is an exchange that runs on many tools, many permissions, and facts that keep changing. We built sho to keep one story current across all of it, and it’s what made our agents actually useful.
Volume: SaucerSwap’s public stats API, all-time, $6.76B.
Same question. Two people. Each gets the answer they’re allowed to see.
When does the pilot start?
The pilot starts Oct 6. It was September until the August decision moved it to October, and the September reminder confirms the date.
When does the pilot start?
The pilot starts in September, from a Discord message in July. A decision I can’t see may change this.
From our benchmark
Read the full answerThere’s a decision I can’t see that may or may not have changed it. […] If the underlying date has moved and you restate it […], you compound the exposure.
list_sources, list_source_fields, search_source_text, read_source, get_source_artifact, get_linked_company_context, find_entities and read_versions, plus reader_status, naming the build that answered. Search takes a question or a phrase: a question is searched by its content terms, a phrase is tried exactly first, and match: literal asks for the exact phrase only. It is never a model prompt. Every answer cites its source down to the exact bytes.search_source_text, trimmed{ "name": "search_source_text","arguments": { "query": "export fix", "limit": 5 } }{ "state": "complete","items": [{"source": { "sourceKind": "discord", "sourceVersion": "…","contentSha256": "…", "businessEffect": "unverified" },"matches": [{ "text": "we will ship the export fix on Friday","citation": { "fieldKey": "content","startByte": 120, "endByte": 157 } }] }] }Context stops being something people go looking for. It comes to the work.
Search finds pages that match your words. sho is for what they mean together: what’s settled, who disagrees, and what changed since.
Because the most recent message is only the ending. What it means depends on everything before it: what was tried, who disagreed, what changed, and why. That story is spread across every tool and every person, and nobody holds all of it. sho does.
Your agents keep doing the work. sho connects through a single MCP server and gives them your company’s story, with a source for every answer. See how it works
sho reads Discord, GitHub, Linear, Gmail, Google Calendar, Google Drive, local files, iMessage, Telegram and WhatsApp. Your agents reach all of it through one MCP server.
It stays on the machine that runs sho, in a local database reached only over loopback. Searching doesn’t send your text to a model provider. Each capture follows your retention policy, and a deleted message stops being served.
Against graded test sets, with the bar to pass set before each run. How we test