Thesis is a graph-based research workspace from Synthetic Sciences. Every claim, experiment, source, and decision lives in an explicit, auditable structure: a graph of research nodes, an append-only research log, indexed knowledge sources, tracked artifacts, and a compute approval boundary that keeps humans in control of what runs and what spends.
Looking for the terminal coding agent? That’s Synsci CLI, a separate product. Thesis is the research workspace; Synsci CLI is the agent you run in your shell. They pair via MCP if you use both.

​
Who Thesis is for

Thesis is built for people running research programs where losing context is expensive.
  • AI researchers comparing methods, prompts, training runs, or evaluations across many experiments.
  • Scientists turning papers, datasets, notebooks, and results into an audit trail they can branch and reproduce.
  • Agent builders who need a real project memory layer rather than fuzzy chat-based RAG.
  • Teams that want agents to plan and execute work without handing over control of budget and research direction.

​
The five core pieces

Thesis combines five capabilities that normally live in separate tools.
  • Research graph. A DAG of staged and committed nodes (hypotheses, experiments, insights, branches, merges). Committed nodes form durable history; new work branches from them instead of overwriting them.
  • Agent harness. A Thesis-native ReAct agent that hydrates from graph state, searches prior work, stages experiments, writes the research log, and spawns approved sub-agents. It does not rely on chat history as its source of truth.
  • Knowledge layer. Unified search across your Thesis graph and external sources indexed through Nia (papers, repositories, documentation, datasets) from one interface.
  • Artifact and filesystem layer. Node artifacts, project volume files, research reports, sub-agent working directories, asset cards, uploads, previews, export, and import. Evidence stays attached to the nodes that produced it.
  • MCP server. 86 structured tools at /mcp-server over JSON-RPC 2.0, available to any MCP-compatible host.

​
Connect Thesis to your agent host

The @synsci/thesis npm helper writes the configuration for every detected host in one command: 
npx --yes @synsci/thesis setup
Supported hosts: Claude Code, Codex, Cursor, OpenCode, Pi, Hermes Agent, OpenClaw, and Synsci CLI.
The @synsci/thesis helper is not the same as @synsci/cli. @synsci/thesis writes Thesis MCP configuration into agent hosts. @synsci/cli is the Synsci CLI terminal agent itself.

​
What makes Thesis different

Research state in Thesis is inspectable the same way source code is inspectable. Every graph mutation happens through a typed, auditable tool, not a hidden PM agent guessing your intent from conversation. Compute-spending sub-agents require explicit approval before they start. Committed nodes cannot be silently rewritten. The research log is append-only. The goal is simple: make research work you can find, branch, reproduce, and hand off, months after the original work happened.
You can start using the Thesis web app without any additional setup. The MCP integration is optional and adds the ability for external agent hosts to read and write your research graph.