Private beta · Research AI

The AI forgets everything
between sessions.
Your research shouldn't.

Scipeer builds a persistent knowledge graph from your work. Every new AI session starts knowing what you've done, what you've tried, and why.

SCROLL TO SEE HOW IT WORKS
01 / 09
01 / 09

The project starts.

Dr. Zhang connects the instruments, the team, the tools, context that's usually scattered across emails and memory.

Every node is typed. Every node has provenance: who added it, when, why.

Provenance & versioning

Every decision, traced.

Human intuition and AI reasoning, side by side. Every action leaves a permanent record, who did it, when, and why.

Researcher Scipeer agent
Dr. Zhang09:14
created project IL-2 / Treg Study
"Investigating biphasic IL-2 dose-response in Treg populations"
Dr. Zhang14:32
logged experiment Flow Cytometry #7, 24-plex panel, 847 MB FCS
"LSRFortessa run, gated on CD4+ Treg populations"
Scipeer14:44
fetched and linked 3 PubMed papers, IL-2 + Treg + dose-response
auto-linked to Experiment #7 · every claim cited to exact source passage
Dr. Zhang14:51
proposed Hypothesis 1, IL-2 dose-response is biphasic in Tregs
"I've seen this pattern in the pilot data, needs a formal titration test"
Scipeer14:52
linked supporting evidence to Hypothesis 1
GSE206999 · Sadlack et al. 2013 · Fontenot et al. 2005, cited to exact passages
Scipeer15:03
proposed Hypothesis 2, inferred from clustering residuals in analysis.ipynb
unexplained variance in CD25-high subpopulation · confidence: moderate
Dr. Zhang15:06
approved Hypothesis 2
"Worth tracking, revisit after next run"
Scipeer15:08
proposed experiment Titration run, 6 IL-2 concentrations, LSRFortessa
to resolve biphasic window in Hypothesis 1 · awaiting PI approval before anything runs
Dr. Zhang
Pending review, titration run awaiting approval
nothing executes until you say so

Under the graph.

Scipeer is a reasoning layer, not another tool. It connects to where your work already lives and makes every AI session start fully informed.

Graph store

PostgreSQL + pgvector. Every entity, relationship, and provenance record is queryable. Semantic search over embeddings, exact query over structure.

  • Typed nodes with full provenance (actor, timestamp, reason)
  • Versioned entities, data files, notebooks, hypotheses
  • Semantic similarity search at query time

MCP tools for Claude

Scipeer exposes 9 MCP tools. When you open Claude Code in your project directory, it already has the full graph in context.

get_context add_experiment add_hypothesis link_literature search_graph propose_action get_timeline export_report get_provenance

Connectors

Scipeer ingests from wherever your research lives. Domain-agnostic, works for wet labs, AI research, physics, computational biology.

  • Any LIMS, openBIS, SampleDB, Chemotion, Benchling, Labguru
  • Jupyter notebooks with executed outputs
  • Semantic Scholar + PubMed literature pipeline
  • Raw instrument files, FCS, HDF5, CSV
  • REST APIs and generic CSV connectors

Staged autonomy

The agent reasons and proposes, it cannot act. Every action that touches the physical world produces a proposal node. A human approves or rejects. That decision is recorded.

  • Proposal → Review → Approve/Reject → Execute
  • Approval recorded as a typed edge with timestamp
  • No silent writes to external systems in v1

Apply for early access.

We're onboarding a small cohort to shape the product. If you've hit the "AI forgets everything" problem and want to help fix it, we want to hear from you.