THE KNOWLEDGE REACTOR / 001

Scattered sources.
One potential.

The reactor begins as separate components.
Scroll to build a connected company core.

01 / SCATTERED COMPONENTS00%

↓ CONTINUE ASSEMBLYSCROLL BACK TO REVERSE ↗

Connects to the tools your company already runs — and anything that speaks MCP or exposes an API.

Your company already knows the answer. It’s just scattered.

Agents are blind — or see too much.
Without a memory layer, AI agents either can’t reach company knowledge or reach far more than the person asking should.
Nobody knows what was true then.
Plans change. Without history, you can’t tell what was known when a decision was made — or which fact replaced it.
An answer without a source is a guess.
If you can’t check where it came from, you can’t act on it, defend it, or hand it to an auditor.

How it works

From scattered to one brain,
in three steps.

01 Connect
Plug in your tools and your agents.
Slack, Drive, GitHub, Jira and the rest connect alongside your agents over MCP. Every source keeps its own permissions.
02 Remember
It builds a living memory.
People, projects, decisions and commitments are linked into one graph. Every fact records when it was true and when it was learned — nothing is silently overwritten.
03 Answer
Ask anything. Get the source.
Answers are trimmed to what the asker may see before the model reads a word, and every claim links back to its evidence.
I’m new. Who owns Billing v2?
- 1Engineering handbookNotion
- 2#platform launch threadSlack
- Withheld
Scoped to your access

Try it

Ask the way your team asks.

Change who’s asking, or when. The answer changes — and so does the evidence behind it.

slashh · AskSample data

Viewing as

As of

I’m new. Who owns Billing v2?

Answer

Maya Chen owns Billing v2 on the Platform team. Jules Rivera handles connector reviews. The engineering handbook and the launch thread agree.

Scoped to a team member · current facts

Your agents get the same answer, with the same limits, over MCP.

Why Slashh

Each claim links to the message, document or decision behind it. Check it in one click.
Permissions apply before the model reads a word. Same question, different people, different answers.
Facts are replaced, never erased. Ask what you knew in March and get March’s answer.
Claude, Cursor, LangGraph, CrewAI — any MCP host gets the same memory, acting for a person and never seeing more than they could.
Every agent call lands in an append-only, hash-chained log your auditors can export.
When two sources disagree, you see both — never a quiet guess.

The launch moved to 24 May¹ so the onboarding review could finish before sign-off².

1Launch plan, revision 3Notion, updated 12 May2Product review notesCalendar, 12 May

Dana, finance lead

Q3 pricing decisionNorthwind renewal notesPlatform roadmapBoard memo, MarchHiring plan

Sam, contractor

WithheldNorthwind renewal notesPlatform roadmapWithheldHiring plan

1 Mar14 MarToday

Billing v2 owner: Arjun PatelBilling v2 owner: Maya Chen

memory_ask("Why did the launch move?")answer, 2 citations, acting for maya@companywho_knows("billing migration") → Maya Chen, Jules Rivera

a41fmemory_ask2 docs read, 1 withheld9c0ewho_knows3 docs reade7b2decisions1 doc read51d8entity_profile4 docs read

Product roadmapSSO ships in Q3≠Acme account callSSO ships in Q2

Use cases

The questions your team asks every day.

Onboarding, asked by a new engineer, week oneThe owner, the reviewer, and the handbook page that says so.

For developers

One config block.
Every agent remembers.

Point Claude, Cursor or any MCP host at Slashh’s hosted endpoint. Every tool returns cited answers, trimmed to what the human behind the agent is allowed to see.

memory_ask
memory_search
who_knows
entity_profile
decisions
glossary_define
timemachine_diff

Read the developer guide

mcp.jsonMCP · HTTP

1{
2  "mcpServers": {
3    "slashh": {
4      "type": "http",
5      "url": "https://YOUR-SLASHH-API/mcp",
6      "headers": {
7        "x-api-key": "REPLACE_WITH_AGENT_KEY"
8      }
9    }
10  }
11}

The difference

Not a chatbot bolted
onto your apps.

Most AI search runs documents through a vector index and hopes for the best. Slashh builds a permission-aware, time-aware model of your company first — then answers from it.

	Slashh	Vector DB / RAG	MCP file access	Wiki
Cites the source behind every answer	Always	Sometimes	No	Manually
Answers “what did we know then?”	Any date	No	No	Page history
Applies permissions before the model reads	Fail-closed	No	File-level	Page-level
Understands people, decisions and commitments	Yes	No	No	Manually
Audit trail of what every agent saw	Hash-chained	No	No	No

Security

Follow one question through Slashh.

Before any model reads a word, the request is tied to a person, trimmed to what they may see, answered with evidence and written to a log that can’t be quietly edited.

How Slashh keeps data safe

Fail-closed by default
A source the asker can’t see never reaches the model — not its title, not a snippet, not a count.
Agents act for people
Every agent acts on behalf of a human, with keys hashed and scope that can only narrow.
A trail you can prove
Every call is attributed in an append-only, hash-chained log, exportable from the SOC2 Evidence Room.
Runs where your data lives
Self-host or air-gap it. MIT-licensed open core, and it runs offline with no API key.

Give your company a brain.

Bring your team’s hardest question. We’ll show you what Slashh can connect — and what it will never let slip.

Book a demo Explore the product

Scattered sources.One potential.

Your company already knows the answer. It’s just scattered.

Agents are blind — or see too much.

Nobody knows what was true then.

An answer without a source is a guess.

From scattered to one brain,in three steps.

Plug in your tools and your agents.

It builds a living memory.

Ask anything. Get the source.

Ask the way your team asks.

Why Slashh: memory you can trust

Every answer, cited.

Sees only what you can.

Remembers what was true when.

Built for agents.

A trail you can prove.

Spots contradictions.

The questions your team asks every day.

One config block.Every agent remembers.

Not a chatbot boltedonto your apps.

Follow one question through Slashh.

Fail-closed by default

Agents act for people

A trail you can prove

Runs where your data lives

Give your company a brain.

Scattered sources.
One potential.

From scattered to one brain,
in three steps.

One config block.
Every agent remembers.

Not a chatbot bolted
onto your apps.