What is the Neural Link?

The Neural Link is a bi-directional bridge that connects AI agents directly to the Neo.mjs runtime. It lets agents inspect the Scene Graph, component state, event listeners, computed styles, and DOM rectangles, and mutate the running application in real time.

Why is Neo.mjs called an Application Engine instead of a framework?

Neo.mjs maintains persistent application objects in a worker-backed Scene Graph instead of compiling application state away into ephemeral DOM nodes. That architecture enables multi-window orchestration, runtime permutation, and deep AI introspection.

What is Context Engineering?

Context Engineering shapes the information and tool environment around AI agents. Neo.mjs implements it through Knowledge Base, Memory Core, GitHub Workflow, and Neural Link MCP servers for frontier harnesses, plus a File System MCP server for internal Neo.ai.Agent local loops.

What is the Neo.mjs Agent OS?

The Neo.mjs Agent OS is the repository Brain: source code and services for Memory Core, Knowledge Base, Active Hybrid GraphRAG, DreamService, Golden Path synthesis, A2A coordination, and Neural Link tooling.

Frontmatter

id	10057
title	Knowledge Base: Add PullRequestSource for PR Conversation Embeddings
state	Closed
labels	enhancementai
assignees	tobiu
createdAt	Apr 18, 2026, 12:48 AM
updatedAt	Apr 18, 2026, 8:44 PM
githubUrl	https://github.com/neomjs/neo/issues/10057
author	tobiu
commentsCount	1
parentIssue	9999
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
blockedBy	[]
blocking	[]
closedAt	Apr 18, 2026, 8:44 PM

Knowledge Base: Add PullRequestSource for PR Conversation Embeddings

Closed v13.0.0/archive-v13-0-0-chunk-4 enhancementai

tobiu commented on Apr 18, 2026, 12:48 AM

Intent / The Why

The Knowledge Base indexes content from six source providers: ApiSource, LearningSource, ReleaseNotesSource, TicketSource, DiscussionSource, and TestSource. The GitHub Workflow server syncs PR conversations to local markdown files at resources/content/pulls/ (~280 files), but no PullRequestSource exists to embed them into the vector database.

This means:

ask_knowledge_base(query='how was the grid selection refactored?') can find the ticket that described the intent, but not the PR that documents the actual implementation, review feedback, and architectural corrections made during execution.
The PR body contains the richest signal: diff summary, reviewer scores, scope-creep warnings, and follow-up ticket links. Without embeddings, this knowledge is invisible to semantic search.

The Q&A Signal Pattern

The old workflow (pre-Swarm) put both problem and solution in ticket comments. With PRs, the signal split: ticket = problem, PR = solution. If the KB only indexes tickets, agents can find what to do but not how it was done before. This asymmetry degrades RAG quality over time as more work flows through PRs.

Scope of Work

New Source Provider

Create ai/mcp/server/knowledge-base/source/PullRequestSource.mjs following the DiscussionSource.mjs pattern:

Scan resources/content/pulls/*.md
Parse YAML frontmatter (number, title, author, state, mergedAt, head, base, url)
Chunk the PR body + conversation into embeddable segments
Tag each chunk with type: 'pull' metadata for filtered queries

KB Query Integration

The query_documents tool already supports type filtering. After adding PullRequestSource, queries with type='pull' should return PR content. No tool changes needed — just the source provider.

Key Files

File	Action
`ai/mcp/server/knowledge-base/source/PullRequestSource.mjs`	[NEW] Source provider
`ai/mcp/server/knowledge-base/config.mjs`	[MODIFY] Register new source

Reference Pattern

ai/mcp/server/knowledge-base/source/DiscussionSource.mjs — this is the closest analog. It scans resources/content/discussions/, parses frontmatter, and chunks content. PullRequestSource should follow identical patterns with PR-specific metadata.

Acceptance Criteria

PullRequestSource extracts and embeds PR markdown files
query_documents(query='grid refactoring', type='pull') returns relevant PR chunks
ask_knowledge_base(query='how was issue #N resolved?') can find the corresponding PR body
Existing source providers unaffected (no regressions)

Origin Session ID: 144326a4-c1e8-4eee-80e4-b984f3c79ddc

tobiu added the enhancement label on Apr 18, 2026, 12:48 AM

tobiu added the ai label on Apr 18, 2026, 12:48 AM

tobiu assigned to @tobiu on Apr 18, 2026, 8:10 PM

tobiu Apr 18, 2026, 8:11 PM

Input from Claude Opus 4.7 (Claude Code):

❋ Graph wiring (ticket-intake)

Parent: #9999 — fits the "KB knows more about project work history" framing. (Originally #9999 §3 scoped PRs under the Macro DB plan; that vector-heavy direction was abandoned due to semantic dilution. This ticket delivers the same goal via the conventional chunked-source pattern instead.)

Related, not parent: #10030 — Concept Ontology is a different architectural layer (deterministic JSONL graph, not vectors). PR embeddings are complementary: concepts tell you what; PR embeddings surface how things got done.

Hypothesis vs root cause (flagged during intake)

The ticket prescribes "chunk the PR body + conversation into embeddable segments", but the actual DiscussionSource.mjs / TicketSource.mjs pattern is one chunk per markdown file — downstream chunking happens in DatabaseService. I will follow the established pattern (~40 LOC, mirror of DiscussionSource) rather than over-engineering per-section splitting.

Minor out-of-scope observations (not bundling into this PR)

TicketSource.mjs:14-17 JSDoc says .github/ISSUE_ARCHIVE but code reads resources/content/issues + issue-archive. Stale comment.

Frontmatter metadata is embedded in content rather than extracted across all source providers. Limits state='merged' style filters; fine for semantic search today.

Proceeding with branch agent/10057-pull-request-source from origin/dev.

tobiu added parent issue #9999 on Apr 18, 2026, 8:11 PM

tobiu referenced in commit 01f6a0f - "feat(kb): add PullRequestSource for PR conversation embeddings (#10057)" on Apr 18, 2026, 8:14 PM

tobiu cross-referenced by PR #10066 on Apr 18, 2026, 8:15 PM

tobiu referenced in commit 8e47e5f - "test(kb): move PullRequestSource spec to canonical unit/ path (#10057)" on Apr 18, 2026, 8:23 PM

tobiu referenced in commit 4e248dc - "test(kb): exercise PullRequestSource singleton against mock fs (#10057)" on Apr 18, 2026, 8:37 PM

tobiu closed this issue on Apr 18, 2026, 8:44 PM

tobiu referenced in commit fde93fe - "feat(kb): add PullRequestSource for PR conversation embeddings (#10057) (#10066) on Apr 18, 2026, 8:44 PM

tobiu cross-referenced by #9748 on Apr 20, 2026, 1:13 PM