GitHubGitHub RepoNeeds Review

Tencent/WeKnora

WeKnora combines self-hosted RAG, agent tools, durable sandboxes, memory, and auto-generated wikis, making it a broad reference stack for document-grounded agents.

GitHub · Trending today

Open Source Open Markdown Open JSON

Source Summary

WeKnora is an open-source document knowledge stack with RAG Q&A, a ReAct agent, and an agent-maintained wiki. Version **v0.8.0** adds persistent Docker, E2B, and Cube skill sandboxes, cross-session memory, and a tenant skill catalog; the project lists **20+ model providers**.

Practical Implication

Study it as an integrated reference for document agents: retrieval, MCP and skill scoping, editable chunks, revision history, observability, RBAC, and private deployment are handled in one system. Its modular pipeline also permits swapping models, vector stores, and storage.

Agent-Ready Context

WeKnora is an open-source document knowledge stack with RAG Q&A, a ReAct agent, and an agent-maintained wiki. Version **v0.8.0** adds persistent Docker, E2B, and Cube skill sandboxes, cross-session memory, and a tenant skill catalog; the project lists **20+ model providers**.

Study it as an integrated reference for document agents: retrieval, MCP and skill scoping, editable chunks, revision history, observability, RBAC, and private deployment are handled in one system. Its modular pipeline also permits swapping models, vector stores, and storage.

The breadth creates operational surface area, and the material provides no independent quality or scale benchmark. Several headline capabilities are recent, while the local host-process sandbox was removed and Docker is opt-in.

Connected Context · Feed7 Judgment

WeKnora turns the candidates’ general memory, skill, retrieval, and governance patterns into one document-agent stack. It confirms the value of a shared substrate, but also concentrates deployment, isolation, tenancy, and upgrade concerns in one system; its recent sandbox and memory features should therefore be evaluated rather than assumed mature.

The Era of Compound Engineering — Kieran Klaassen, Every/CoraWeKnora’s cross-session memory and revision history provide infrastructure for the accumulated, curated context advocated by compound engineering, while not establishing that the stored material remains useful or current.affaan-m/ECCBoth combine skills, memory, and harness controls, but WeKnora narrows them to document-grounded agents whereas ECC offers a broad cross-client configuration catalog; each creates a similar selective-adoption and audit burden.A Taxonomy of Cognitive Capability Gaps in Generative and Agentic AIWeKnora implements several supports identified by the capability-gap taxonomy, including durable state, monitoring, and controlled access, but the taxonomy reinforces that architecture alone does not validate long-horizon reliability.melgarafael/DeskcommCRMBoth integrate tenant-scoped retrieval, memory, skills, MCP, RBAC, and auditing; DeskcommCRM applies that substrate to operational CRM actions, while WeKnora centers document knowledge and editable evidence.

Context Map

agentresearchdata#harness-engineering#agent-memory#skills

Uncertainty

The breadth creates operational surface area, and the material provides no independent quality or scale benchmark. Several headline capabilities are recent, while the local host-process sandbox was removed and Docker is opt-in.