How agentscope-ai/reme works
ReMe is a local-first, file-native knowledge base for AI agents: users and agents write Markdown memories, ReMe indexes them, then retrieves relevant passages and linked context. The central runtime configures components and exposes workflow jobs through HTTP/MCP or other interfaces. Separate workflows evolve conversations and resources into durable notes, and proactively consolidate knowledge. Studio provides a workspace for interacting with those files. The README documents these capabilities; sampled source confirms application wiring, component lookup, and HTTP job routing, but several domain-stage connections are documented rather than source-verified.
Main entry points: http_service.py HTTP and MCP service · local_file_store.py Markdown workspace · application.py Application runtime · search.py Hybrid memory search
Components 20
Runtime and jobs
- Application runtime
reme/application.py - Component registry
reme/components/component_registry.py - Workflow steps
reme/steps/base_step.py - Requests and results
reme/schema/request.py
Memory lifecycle
- Markdown file operations
reme/steps/file_io/read.py - Markdown workspace
reme/components/file_store/local_file_store.py - Conversation capture
reme/steps/evolve/auto_memory.py - Resource ingestion
reme/steps/evolve/auto_resource.py - Daily notes
reme/steps/file_io/daily_write.py - Knowledge consolidation
reme/steps/evolve/dream/integrate.py - Proactive discovery
reme/steps/evolve/proactive/proactive.py
Index and recall
- Search indexing
reme/steps/index/reindex.py - Hybrid memory search
reme/steps/index/search.py - Wikilink expansion
reme/utils/link_expansion.py
User interfaces
- HTTP and MCP service
reme/components/service/http_service.py - HTTP client
reme/components/client/http_client.py - Studio workspace
reme_studio/app/workspace.tsx
Agent ecosystem
- Agent integrations
integrations - Community plugins
plugins
Other components
- User or agent
Connections 21
- User or agent → HTTP and MCP service: submits actions
README.md - HTTP and MCP service → Application runtime: routes jobs
reme/components/service/http_service.py - Application runtime → Component registry: resolves backends
reme/application.py - Application runtime → HTTP and MCP service: initializes service
reme/components/service/http_service.py - HTTP and MCP service → Requests and results: accepts requests
reme/components/service/http_service.py - HTTP client → HTTP and MCP service: sends actions
- Application runtime → Workflow steps: runs configured jobs
- User or agent → Markdown file operations: reads and writes
README.md - Markdown file operations → Markdown workspace: stores Markdown
README.md - Conversation capture → Daily notes: creates notes
- Resource ingestion → Markdown workspace: preserves sources
README.md - Daily notes → Markdown workspace: writes notes
- Knowledge consolidation → Markdown workspace: refines knowledge
README.md - Proactive discovery → Markdown workspace: discovers knowledge
- Markdown workspace → Search indexing: indexes files
README.md - Search indexing → Hybrid memory search: supports retrieval
- Hybrid memory search → Wikilink expansion: expands context
README.md - User or agent → Hybrid memory search: queries memories
README.md - User or agent → Studio workspace: uses workspace
- Agent integrations → HTTP and MCP service: shares workspace
- Community plugins → Workflow steps: extends workflows
plugins/beam-judge/src/judge_beam/llm_judge.py