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.

Neo.mjs - Self-Evolving Software Organism

Frontmatter

title	fix(ai): chunk oversized kb sources before embedding (#14000)
author	neo-gpt
state	Merged
createdAt	Jun 25, 2026, 1:22 PM
updatedAt	Jun 25, 2026, 3:44 PM
closedAt	Jun 25, 2026, 3:20 PM
mergedAt	Jun 25, 2026, 3:20 PM
branches	`dev` ← `codex/14000-kb-oversized-source-chunking`
url	https://github.com/neomjs/neo/pull/14003
contentTrust
projected
quarantined	0
signals	[]

Merged

neo-gpt commented on Jun 25, 2026, 1:22 PM

Resolves #14000

Adds deterministic oversized-source chunking inside KnowledgeBaseIngestionService before the final embedding guardrail. Recoverable raw fallback and parser-produced text chunks are split into embedding-safe sub-chunks, keep trace metadata, and receive stable rehashed IDs; unsplittable chunks still use the existing bounded KB_INGEST_INPUT_SIZE_EXCEEDED diagnostic. VectorService remains the final refusal guardrail, and no provider limits are raised.

Evidence: L2 (unit-level ingestion path with mocked VectorService boundary and deterministic ID assertions) -> L2 required (pre-provider KB ingestion behavior). No residuals.

Related: #13999 Related: #13930 Related: #13929

Deltas from ticket

No response schema change was needed: the existing summary fields already distinguish embedded chunks from final skipped chunks. Split metadata is stored on generated chunk records as oversizedSplit, oversizedSplitIndex, oversizedSplitTotal, oversizedSplitCharStart, and oversizedSplitCharEnd.

Test Evidence

npm run agent-preflight -- ai/services/knowledge-base/KnowledgeBaseIngestionService.mjs test/playwright/unit/ai/services/knowledge-base/KnowledgeBaseIngestionService.spec.mjs passed.
npm run test-unit -- test/playwright/unit/ai/services/knowledge-base/KnowledgeBaseIngestionService.spec.mjs passed: 37/37.
git diff --check passed.
Freshness check passed after rebase: merge-base HEAD origin/dev == origin/dev at f97d1562a63272c86fb44e71e055111b19794ad2.

Post-Merge Validation

Run a KB sync including resources/content/issues/chunk-2/issue-12065.md and verify rows exist for that sourcePath with oversizedSplit metadata instead of a zero-vector skip.

Commits

2603f918cc - fix(ai): chunk oversized kb sources before embedding (#14000)

Evolution

This spun out of the #13999 backup investigation after the operator's log sample proved a separate KB completeness gap: the KB store can remain exportable while an oversized source is missing entirely because the prior safety behavior skipped instead of splitting. This PR keeps the safety invariant and recovers vectors for split-safe sources.

Authored by Euclid (GPT-5, Codex Desktop). Session f4d00667-a65a-4285-83f5-6761f3aea394.

neo-gpt commented on Jun 25, 2026, 3:02 PM

Author-side ADR-0019 correction: the challenged test setup was a real B4 violation shape. The ingestion spec was mutating the shared Provider singleton via memoryConfig/aiConfig to force local embedding caps. I pushed 6f306053b7 to replace those writes with a test-local resolveEmbeddingInputGuardrail override restored through the existing singleton seam.\n\nValidation after the fix:\n- npm run test-unit -- test/playwright/unit/ai/services/knowledge-base/KnowledgeBaseIngestionService.spec.mjs passed, 37/37.\n- npm run agent-preflight -- test/playwright/unit/ai/services/knowledge-base/KnowledgeBaseIngestionService.spec.mjs passed.\n- git diff --check passed.\n- rg found no remaining memoryConfig/localModels.embedding mutation in the touched ingestion spec slice.

tobiu APPROVED reviewed on Jun 25, 2026, 3:19 PM

No review body provided.