Kompl boundary

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

Boundary and evidence

This run examines Kompl as a knowledge system that compiles incoming sources into retained wiki pages and serves those pages to people and agents. The intended use is to establish its source-native operation, including how knowledge is produced, admitted, persisted, maintained and consumed. It does not evaluate Commonplace adoption.

The target class is memory/knowledge/context-engineering system, not an enclosing general-purpose agent runtime. Kompl issues model calls for extraction, drafting and query answering, while its shipped MCP server serves retained pages to an external agent. The boundary kind is whole-system: the target includes the Next.js application and its state-owning callers, Python NLP/model/storage service, shipped n8n workflow, CLI startup maintenance, operator instructions and evaluation paths. The built-in chat is included even though the README calls it replaceable. It consumes retained content and can produce another draft. SRC-1: README.md, app/src/app/api/chat/route.ts, mcp-server/index.ts.

The agent answers questions; Kompl stores and serves the compiled pages. --- README.md

The initial trace establishes a complete shipped loop rather than only a library API:

• Produce and update: onboarding staging and finalization lead through the shipped n8n webhook to the application's compile driver. That driver dispatches ingestion, extraction, resolution, matching, planning, drafting, cross-reference, commit and schema steps. Existing pages and the retained schema are loaded into later drafting calls. SRC-1: app/src/app/api/onboarding/stage/route.ts, app/src/app/api/onboarding/finalize/route.ts, n8n/workflows/session-compile.json, app/src/app/api/compile/run/route.ts, app/src/app/api/compile/draft/route.ts, nlp-service/routers/pipeline.py.
• Admit and persist: compile commit includes automatic/manual approval paths; explicit draft approval uses a shared commit implementation. Database outbox content is flushed through the Python storage service, and startup instrumentation reattempts pending flushes. These are included responsibilities, not external host behavior. SRC-1: app/src/app/api/compile/commit/route.ts, app/src/lib/approve-plan.ts, app/src/lib/flush-pending.ts, app/instrumentation.ts, nlp-service/services/file_store.py.
• Maintain and reload: source deletion/recompilation, wiki lint, import/export, vector maintenance and schema migrations are in scope. The CLI invokes startup lint and backup against the application API. SRC-1: app/src/lib/source-delete.ts, app/src/lib/recompile.ts, app/src/app/api/sources/[source_id]/recompile/route.ts, app/src/app/api/wiki/lint-pass/route.ts, app/src/app/api/import/route.ts, app/src/app/api/export/route.ts, cli/src/startup-tasks.ts, scripts/migrate.py.
• Consume and feed back: the wiki reads retained markdown; MCP calls the wiki's HTTP API; built-in chat retrieves retained pages, calls answer synthesis, persists conversation history and may create a pending query-generated draft. SRC-1: app/src/app/wiki/[page_id]/page.tsx, app/src/app/api/wiki/[page_id]/data/route.ts, app/src/lib/retrieval.ts, app/src/app/api/chat/route.ts, mcp-server/index.ts.

The following implementation excerpts anchor the classification and inclusion of both later update and query-feedback paths. They establish shipped code, not successful execution or answer fidelity.

if existing_content: parts += ["Existing page content (update this, don't rewrite from scratch):", existing_content, ""] --- nlp-service/services/llm_client.py

// 7. Compounding: if answer used 3+ pages, create a pending draft if (pages.length >= 3) { --- app/src/app/api/chat/route.ts

Frozen evidence and limits

The only source is SRC-1, the acquired repository at full commit 17970c969b0c2576bf722840e0e2db3c6416a7aa. Its checkout HEAD was checked against that pin. The analysis cutoff is 2026-10-14, the supplied run date; this is not a claim that the source's last commit occurred on that date. Later commits, live issues, linked but untracked documents and other repositories do not supply evidence. The evidence tier is code-grounded.

External dependencies include Docker/Compose, n8n's upstream runtime, Node/Next.js, FastAPI and local NLP/vector libraries and models, Gemini/DeepSeek provider services, and ingestion services such as Firecrawl and YouTube. Their Kompl-facing calls and configuration belong to the target; their independent internals do not. External MCP clients' scheduling, context assembly, decisions and memory writes are excluded, preventing attribution of an enclosing agent loop to Kompl. SRC-1: docker-compose.yml, README.md, nlp-service/routers/pipeline.py, mcp-server/index.ts.

No target code, tests, examples or model calls were executed. This prevents claims of observed correctness, crash recovery, security or sustained compounding at this revision. The repository's May provider-validation reports are attributed historical evidence, not fresh tests of the reviewed commit. The May 22 Flash report still marks its results pending. The corpus referenced by the validation runners, nlp-service/test_corpus/, is absent from the tracked tree; that access gap prevents reproducing or inspecting those exact input texts within this boundary. Initial inspection of evaluator entry points does not establish causal comparisons or semantic accuracy. SRC-1: validation/2026-05-05-multi-provider-corpus/REPORT.md, validation/2026-05-05-multi-provider-corpus/runner.py, validation/2026-05-22-deepseek-flash/REPORT.md.

Top-level coverage

All top-level tracked entries are classified below. Included means part of the functional target, not fully audited. “Tree/search” identifies initial bounded inspection; available files not yet read in full remain available to later analysts under SRC-1 and are not access gaps. Deliberate exclusions are peripheral to the knowledge loop.

Source register