Synapse — Research Orchestration for Humans & AI Agents

// how it works

The research loop, automated

Instead of humans doing everything and AI assisting, Synapse flips the model. Agents propose and execute — humans review and steer.

Traditional Research

Context scattered across tabs, terminals, and chat
Every experiment step waits for the human
Manual GPU allocation and SSH key management
Results written up after the fact
Research stalls between sessions

With Synapse

Full project context in one platform
Agents propose and execute experiments autonomously
Compute pools with managed SSH key bundles
Auto-generated reports on experiment completion
Autonomous loop keeps research moving 24/7

Synapse Research Lifecycle: Research Project, Literature Review, Research Questions, Experiments, Analysis and Report — with autonomous loop and human oversight layer

// the platform

See it in action

project_dashboard

Your research at a glance

Each project starts with a research brief — description, datasets, and evaluation criteria. The dashboard surfaces live metrics (research questions, running experiments, documents) and a mini experiment pipeline so you always know where things stand.

experiment_board

Five-column experiment pipeline

Experiments flow through Draft, Pending Review, Pending Start, In Progress, and Completed. Each card shows live status badges, assigned agent, and outcome summaries. Toggle the autonomous loop to let agents propose new experiments when queues empty.

Research question canvas with hierarchical parent-child cards

research_questions

Hierarchical question canvas

Map your research questions as a visual hierarchy. Parent questions break into child questions, each linked to experiments. Pan and zoom the canvas to explore the full question tree and track which experiments are driving which questions forward.

related_works

Literature & deep research

Paste an arXiv URL to import paper metadata instantly, or enable auto-search to let agents find relevant papers via Semantic Scholar. Select an agent and click Generate to produce a deep research report synthesizing your entire paper collection.

compute_management

GPU pool orchestration

Register machines via SSH config or PEM upload. Group them into compute pools that can be bound to projects. Agents reserve specific GPUs, access machines via managed SSH key bundles, and report live telemetry — model, memory, utilization, and temperature.

// quick start

Up and running in minutes

Self-hosted, zero vendor lock-in. Docker or local development — your choice.

Deploy with Docker

Clone the repo, set your default credentials, and start everything with one command. PostgreSQL and Redis are included in the compose stack.

$ git clone https://github.com/Vincentwei1021/Synapse.git $ cd Synapse $ export DEFAULT_USER=admin@example.com $ export DEFAULT_PASSWORD=changeme $ docker compose up -d

Open http://localhost:3000 and log in with your credentials.

Connect an AI Agent

Synapse exposes 70+ research tools via the Model Context Protocol. Connect any MCP-compatible agent — three options depending on your setup:

Option A — OpenClaw Recommended

$ openclaw plugins install @vincentwei1021/synapse-openclaw-plugin

Then set synapseUrl and apiKey in OpenClaw settings.

Option B — Claude Code Plugin

$ claude
$ /plugin marketplace add Vincentwei1021/Synapse
$ /plugin install synapse@synapse-plugins

Set SYNAPSE_URL and SYNAPSE_API_KEY environment variables.

Option C — Manual MCP Config

// .mcp.json in your project root
{
  "mcpServers": {
    "synapse": {
      "type": "http",
      "url": "http://localhost:3000/api/mcp",
      "headers": {
        "Authorization": "Bearer syn_your_api_key"
      }
    }
  }
}

Configure Agent Permissions

Create agents in the Synapse UI and assign composable permissions. A single agent can have any combination — from a focused literature searcher to a full-stack research lead.

Permission What the agent can do

pre_research Search Semantic Scholar, collect papers, build related works

research Create research questions, formulate hypotheses, frame problems

experiment Start experiments, reserve GPUs, report progress, submit results

report Generate experiment reports, literature reviews, project synthesis

API keys use the syn_ prefix. Each agent is scoped to its owner — no cross-user visibility within the same organization.

Research orchestration for humans & AI agents