Skip to content
Network Memory

Your firm's compounding research memory.

Every run feeds a private knowledge graph for your team. Future runs cite prior work. Provenance is preserved. Share a sourced view with a single link — without losing track of who saw what.

Cross-run intelligence is included on Scale. Network Memory grows with every report your team runs.

Used by teams at
Why teams switch

Three problems that disappear.

The problem

Last year's deal memo lives in someone's Downloads folder.

Research is done, then lost. The next associate writes the same memo from scratch — paying again, in hours and in tokens, for work the firm has already done.

With Zyphv

Every run is captured, embedded, and queryable.

Past runs are vectorised and indexed per tenant. New runs pull from prior work first — cheaper, faster, and more consistent than starting clean every time.

The problem

The knowledge graph is in your head.

You know that Acme spun out of NewCo, that the CTO came from Acquired Co, that their 3PL partner is the same one a portfolio company uses. Nobody else does.

With Zyphv

Entities and relationships, extracted automatically.

Companies, people, and relationships are pulled into a knowledge graph as runs complete. Search it. Browse it. Cite it in the next memo.

The problem

Sharing a report = forwarding a PDF and hoping.

Once a memo leaves the firm, you don't know who read it, what they followed up on, or whether the citations are still good.

With Zyphv

Tokenised share links with provenance preserved.

Share a sourced view via a tokenised link. The recipient sees the report, the sources, and the provenance — without an account, and without leaking the rest of your network.

How it works

Four layers, one memory.

Every run is captured, indexed, related, and shareable — by default.

01
Run capture
Every CI, DD, and CRE run is saved with full agent trace and source list.
PostgreSQL · per-tenant schema
02
Cross-run embeddings
Outputs are chunked, embedded, and indexed for similarity search across the team's work.
pgvector · pgai vectorizer
03
Knowledge graph
Companies, people, deals, and relationships are extracted into kg_entities and kg_relations.
graph entities + relations
04
Provenance & share
Cryptographic hashes per agent. Tokenised share links. Audit trail per output.
SHA-256 · tenant isolation
Sample view

What network memory looks like.

A new run reuses prior context. The graph shows what your firm already knows.

Knowledge graph — Acme Robotics
12 prior runs · 184 entities · 412 relations · last touched 4 days ago
2026-06-04

Three of your prior runs touch this entity.

Acme appears in a Q1 CI run on robotics tooling, a Q2 sector map, and a partner-introduction memo. The new run incorporates that context — and credits are discounted accordingly.

3
Linked prior runs
−9
Credits saved on this run
17
Related entities

CEO previously at NewCo. Series B led by Capital A. 3PL partner is shared with Portfolio Co X. Two named competitors overlap with the Q2 sector map.

zyphv.com/share/8a3f…read-onlyprovenance preservedrevocable
Pricing

Network memory is built in.

Every plan captures runs and provenance. Cross-run intelligence is included on Scale; Teams adds shared workspaces.

Starter
$9 / 100 credits

Single user. Runs are captured and searchable.

  • Run history + search
  • Provenance + citations
  • Tokenised share links
  • Knowledge graph (view)
Choose Starter
Scale
$69 / 1,000 credits

Cross-run intelligence and API access. The full memory feature set.

  • Everything in Professional
  • Cross-run intelligence
  • API + webhook integration
  • Tokenised shareable workspaces
Choose Scale

Looking for a shared team workspace? See team plans (Starter $99 / Growth $249 / Enterprise custom).

We don't promise magic. We deliver clarity when it matters most.

Questions

Reasonable answers.

How is the knowledge graph built?
After each run, an extraction step identifies entities (companies, people, deals) and relationships and writes them to the kg_entities and kg_relations tables in your tenant schema. The graph grows as your team runs more reports.
Can other tenants see my data?
No. Tenant isolation is absolute. Every tenant has its own PostgreSQL schema; no cross-tenant queries are ever issued. Files, runs, and graph entities never leave your tenant.
What does "cross-run intelligence" actually do?
When you start a new run, the pipeline searches your tenant's prior runs and uploaded files for relevant context, and either credits prior work or feeds it back to the agents. This typically lowers credit cost and improves consistency.
How does provenance work?
Every agent call writes a SHA-256 hash of input and output to an append-only audit trail. Every report section records which agent produced it, the model used, and the source links. You can independently verify the chain.
Can I share a single section, not the whole report?
Yes. Share links are tokenised and scoped. You can revoke any share link from the dashboard at any time.
Is this on-prem-able?
Not today. The platform is multi-tenant SaaS with strict tenant isolation. Reach out if you have a regulated workload — we are happy to discuss.
More from Zyphv

Explore other intelligence verticals.

Ready when you are.

Start free — 50 credits on us. No card. No noise.

Begin building your memory