topprismwiki

TopprismWiki

Language / 语言: English primary · 中文 README

Enterprise Knowledge Governance Layer for Agentic AI

TopprismWiki is an evidence-governed knowledge layer that turns approved enterprise evidence into traceable formal knowledge for AI agents.

Traditional retrieval helps agents find information. TopprismWiki governs what can become knowledge, why it is trusted, and how it may be published and consumed.

It is implemented as an Agent Skill package and a dependency-free Python reference runner. The long-term direction is an enterprise knowledge operating layer; the public repository currently provides the governed contracts and reference workflow needed to build toward that direction.

Why TopprismWiki

Agentic systems do more than retrieve documents. They summarize, recommend, update records, and take actions. That makes provenance and change control part of the knowledge problem:

• Which source is approved for this decision?
• What evidence supports a fact or relationship?
• How should conflicting statements coexist?
• Who or what may publish a change?
• Can an agent distinguish formal knowledge from a candidate or an unresolved source?

TopprismWiki places a governed publication layer between enterprise sources and agent-facing context:

Enterprise evidence
        |
        v
Source adapters and evidence contracts
        |
        v
Review, entity and relationship governance
        |
        v
Deterministic formal publication
        |
        v
Trusted Knowledge Vault
        |
        v
Query, context and graph projections for AI agents

What it is

TopprismWiki provides:

• evidence contracts for approved sources and adapter-produced records;
• source hashes, evidence anchors, review packages and conflict states;
• canonical entities and typed relationships with separate candidate overlays;
• review-gated, atomic publication into a formal Markdown Vault;
• bounded query, context, graph, batch and diagnostic projections;
• a single public entry Skill, $topprismwiki, with internal child modules.

TopprismWiki is not a vector database, a general-purpose RAG framework, an automatic truth generator, or a bundled WeChat/Office/OCR connector. The public runner deliberately separates built-in governance behavior from deployment-owned adapters and model services.

Enterprise Knowledge Lifecycle

flowchart TB
    S[Enterprise sources] --> A[Source adapters]
    A --> W[Evidence Workspace]
    W --> G[Review and governance]
    G --> C[Deterministic committer]
    C --> V[Trusted Knowledge Vault]
    V --> R[Query, context and graph]
    R --> AG[AI agents]
    W -. conflicts, candidates, status .-> R

The three-layer boundary is intentional:

Co-occurrence, group names, member lists, filenames and brand mentions do not establish formal relationships by themselves.

Core capabilities

See the capability matrix for the complete boundary.

60-second synthetic demo

The public demo uses synthetic chat events only. It verifies the same governed path without exposing private sources:

git clone https://github.com/topprismdata/topprismwiki.git
cd topprismwiki
python3 -m unittest discover -s skills/topprismwiki/tests -v

For the complete end-to-end path, follow the synthetic walkthrough. A clean run currently demonstrates:

2 processing units
11 input messages
1 formalized fact
1 evidence-insufficient unit
1 atomically committed formal page
accepted post-commit validation

Preview creates Workspace evidence and does not write the formal Vault. Update requires an explicit authorization flag and a reviewed decision package.

How it differs from common patterns

This is a comparison of representative architecture patterns, not a benchmark or a claim about every product in a category.

Current evidence and evaluation

The current public package is validated as a governance reference workflow, not as a production-scale knowledge quality benchmark.

• package validation covers eight nested Skill modules;
• the test suite covers authorization, preview isolation, atomic update, rollback, query, graph, retry and diagnostic paths;
• the public safety scan rejects private paths, credentials and unsanitized source examples;
• the synthetic walkthrough verifies preview, reviewed update, validation, query, graph and dashboard behavior.

The evaluation guide defines the next measurement layer: traceability coverage, admission precision, conflict handling, idempotence, rollback correctness and, later, labeled knowledge-quality and agent-task benchmarks. No accuracy, hallucination-reduction or deployment improvement claim is made by this repository today.

Quick start

Read Getting started and run the synthetic walkthrough before connecting real sources.

python3 skills/topprismwiki/scripts/topprismwiki.py init --project /path/to/my-wiki
python3 skills/topprismwiki/scripts/topprismwiki.py doctor --capability core --project /path/to/my-wiki
python3 skills/topprismwiki/scripts/topprismwiki.py preview --source all --project /path/to/my-wiki
python3 skills/topprismwiki/scripts/topprismwiki.py status --project /path/to/my-wiki
python3 skills/topprismwiki/scripts/topprismwiki.py dashboard build --project /path/to/my-wiki

If a command returns blocked or a non-zero exit code, run diagnose first:

python3 skills/topprismwiki/scripts/topprismwiki.py diagnose --project /path/to/my-wiki

Only an explicitly authorized update with a reviewed decision package may write the formal Vault:

python3 skills/topprismwiki/scripts/topprismwiki.py update \
  --source documents --authorized --project /path/to/my-wiki

Documentation

• Concepts
• Product architecture
• Governance model
• Adapter contracts
• Dependencies and agent routing
• Operations and state semantics
• Troubleshooting
• Privacy contract
• Evaluation
• Roadmap
• 中文文档

Privacy and contribution boundaries

The repository contains synthetic contacts, chats, customers, projects, filenames, hashes, paths and timestamps only. Real messages, business documents, Vaults, Workspaces, ledgers, model endpoints and credentials must remain outside Git.

Run the public safety scanner before release:

python3 skills/topprismwiki/scripts/check_public_safety.py .

See SECURITY.md and CONTRIBUTING.md before opening an issue or pull request.

TopPrism status

License

Apache-2.0. See LICENSE.