Nemsy boundary

Type: agentic-system-analyses/types/agentic-system-boundary.md

Boundary and evidence

Nemsy is in scope as a model-dependent personal knowledge system. Its intended use is to turn a user's local source documents and selected conversations into a Markdown Wiki, then use retained pages to answer later questions and propose maintenance. This run covers the whole shipped product, not just the Python agent API. The frozen repository includes its CLI driver, Web backend, browser caller, prompts, storage helpers, configuration and smoke-test definitions (SRC-1: README.md, pyproject.toml, src/nemsy/agent.py, src/nemsy/cli.py, src/nemsy/web.py, frontend/src/views/Chat.tsx, tests/test_smoke.py).

The target class is memory/knowledge/context-engineering system; the boundary kind is whole-system. That classification follows the retained-content loop, rather than the use of “Agent” in project descriptions. Python code loads existing Wiki pages into model context, requests model-generated summaries or insights, writes them into the Wiki, and updates the index. The shipped callers initiate ingest, query, chat, save and lint; they are part of the target rather than external enclosing applications (SRC-1: src/nemsy/agent.py, src/nemsy/cli.py, src/nemsy/web.py, frontend/src/views/FileLibrary.tsx, frontend/src/views/Chat.tsx).

Minimum source passages supporting the classification and loop:

wiki_context = _load_wiki_context(max_files=50 if wide else 10) --- src/nemsy/agent.py

file_path.write_text(summary_content, encoding="utf-8")

_update_index(source_title, filename, f"来自 {source_title} 的摘要") --- src/nemsy/agent.py

for f in other_files[:max_files]: rel = f.relative_to(settings.vault.wiki_path) content = f.read_text(encoding="utf-8") context_parts.append(f"=== {rel} ===\n{content}")

--- src/nemsy/agent.py

Entry-point tracing started at the three script registrations in pyproject.toml: nemsy.cli:main, nemsy.cli:smoke and nemsy.cli:web. main dispatches commands and defaults to the chat loop. The CLI holds conversation history and a session Wiki-context snapshot, and routes ingest/save/archive requests to agent operations. The Web command starts src/nemsy/web.py, whose routes also implement their own chat/query prompt construction and call the LLM wrapper directly. Browser Chat.tsx supplies history and explicitly requests insight saving; FileLibrary.tsx requests source ingestion and reads retained Wiki files. These paths must be assessed separately: inclusion does not assume CLI/Web behavioral parity (SRC-1: pyproject.toml, src/nemsy/cli.py, src/nemsy/web.py, frontend/src/views/Chat.tsx, frontend/src/views/FileLibrary.tsx).

Included state responsibilities are source-file collection and directory _index.md guidance; ingest status records; model-generated source summaries, query archives and conversation insights; Wiki index and operation logs; token-use logs; later Wiki-context loading; browser file and citation consumption; model-based lint suggestions; and the shipped smoke-test assertions. Page-schema text and task prompts are product instructions, unlike the repository's developer-assistant rules (SRC-1: src/nemsy/vault.py, src/nemsy/schema.py, src/nemsy/agent.py, config/_index.example.md, .catpaw/rules).

Repository coverage

Every tracked top-level file or directory is classified below. “Excluded” means deliberate functional exclusion, not unavailable evidence. Initial inspection is specified rather than implying every blob was read.

The revision is 2338bab0aab96a91efb90a80249f33510b630260, with applicability cutoff 2026-11-09. The acquired checkout's HEAD matched that revision and its short Git status was empty when inspected. The overall evidence tier is code-grounded: implementation and declared intent are available, but this run supplies no target execution or causal experiment.

External dependencies are the model service reached through the Anthropic SDK, Python/Web/browser libraries, the host filesystem, and user-supplied documents and directory guidance. The configured service is DeepSeek's Anthropic-compatible endpoint; model-service internals and provider-side caching are outside this repository boundary. Obsidian is an optional environment for viewing the Markdown, not a prerequisite established by the product instructions (SRC-1: config/settings.toml, src/nemsy/llm.py, pyproject.toml, README.md).

No user Vault, .nemsy runtime records, live service responses or test execution results were supplied. These unavailable operational inputs prevent conclusions about a real collection's quality, model compliance, effective cache hits, successful deployment, measured costs or long-term knowledge improvement. They do not prevent source analysis of the shipped retention-and-consumption paths. Dependency implementations and future revisions are excluded; the run cannot establish provider guarantees, SDK internals or behavior after the pinned commit. Available but initially uninspected files in this same commit remain available to later source inspection under SRC-1; they are not access gaps.

Source register