| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 10 天前 | ||
| 10 天前 | ||
| 10 天前 | ||
| 13 天前 | ||
| 10 天前 |
ScienceDiscovery Documentation
This is the complete English documentation set.
Tutorials
- Quick start — install, start the stack, configure a model, and complete a first agent task.
- Evolve a solution — run a program-evolution search end to end and judge whether the improvement is real.
How-to guides
-
Configure custom MCP servers: local/remote connections, secret editing, OAuth, Inspector and Session tool selection.
-
Deployment — local mode, Docker, and building a single-file binary from source.
-
Configure the network proxy — add a proxy on the settings page and choose a policy for LLM, web, and MCP traffic.
-
Literature research case guide — run an end-to-end cross-database literature survey, from service startup to approvals and results.
-
Run an evolution search — propose a
/evolve-designsearch, size its shards, watch it run, and read the held-out result. -
Install Neo4j and configure ScienceMemory — install an external Neo4j, enable ScienceMemory in system settings, and explore chains in the frontend graph.
Reference
- Configuration — environment variables, default ports, upload/workspace/output quotas, and data layout.
- REST API — internal HTTP API used by the UI: authentication, request/response, and error semantics.
- Runtime behavior and limits — models, settings inheritance, skills, permissions, timeouts, and execution limits.
- Built-in tools — parameters, boundaries, and exposure conditions for model-visible tools.
- Web tools — web search/fetch providers, configuration, permissions, caching, and audit.
- Repository layout — directories, modules, default ports, and data locations.
- Paper Worker — PDF extraction protocol, pipeline, and limits.
- Web frontend — frontend stack, event mapping, and development/test entry points.
Explanation
See the explanation index, which covers the overall architecture, control plane, agent backend, sandbox, Ascend NPU host broker, rate limiting, connectors, MCP tool protocol, network proxy, review and provenance, ScienceMemory, skill progressive disclosure, subagent orchestration, and content-addressable storage.