A living local corpus
Content-addressed ingest captures repository source, including uncommitted working-tree changes, into local SQLite. Watch mode updates what changed.
Grounded code intelligence
Build knowledge.
Use it. Verify it.
CoreStory turns a living codebase into source-linked knowledge that humans and agents can navigate, question, and inspect.
Available
Local ingest, search, watch, graph exploration, source traversal, and read-only agent access.
Local ingest, search, watch, graph exploration, source traversal, and read-only agent access.
Prototype
Trust Engine concepts for evaluating grounding and quality bars.
Trust Engine concepts for evaluating grounding and quality bars.
Proposed
Hosted, federated, portfolio, and enterprise operating models.
Hosted, federated, portfolio, and enterprise operating models.
The product boundary
What runs today
CoreStory is the product. Plot is its internal knowledge-graph and extraction engine. V3 is site and initiative shorthand—not a product-generation claim.
Structure before answers
Explore files, entities, relationships, flows, capabilities, and call paths instead of reducing a codebase to disconnected chunks.
Evidence back to source
Traverse findings and relationships to supporting evidence, then open the source context that earned the claim.
The working loop
How it works
A direct path from repository selection to inspectable source—not an answer box detached from its inputs.
Choose a repository
Ingest
Orient
Search
Explore relationships
Ground claims
Read the source
One knowledge substrate
Product surfaces
Use the same corpus through the surface that fits the work.
Desktop + browser
Console seams and graph, explore, files, and flow views for visual orientation.
CLI
Ingest, refine, search, watch, usage, and serve from the developer workflow.
MCP
Eight read-only tools let agents search, inspect files, and traverse grounded knowledge.
HTTP
A programmatic read surface over the same local corpus and source context.
Measured, not generalized
Evidence, with boundaries
Results from a 34-repository ingestion campaign across 24 languages, totaling 4.7 GB of source. These results demonstrate CoreStory's ingestion capability at scale; they do not establish performance across every language or repository.
Grammar-derived entities
Structural graph extraction across 34 open-source repos in 24 languages.
Source processed
Total source volume ingested and refined through a single-model pipeline.
Languages measured
C, C++, Java, C#, Python, JavaScript, TypeScript, Go, Rust, Clojure, PHP, Fortran, COBOL, Ada, VB.NET, PL/SQL, RPG, ABAP, Delphi, VB6, PowerBuilder, ColdFusion, Oracle Forms.
Known limits: Single-model pipeline (DeepSeek V4 Flash). COBOL, Python, and Rust figures from small repos (2–4 MB). Seven legacy-language results are estimates from entity counts — store.db files were recycled after the run.
Prototype boundary
Trust Engine
An emerging evidence-control and evaluation system for inspecting grounding and testing quality bars—not a production trust score, certification, or guarantee.
Prototype
01Typed claims
02Multi-anchor receipts
03Snapshots and claim plans
04Signed receipts
05“Why” resources
Direction, not commitments
Where this can go
The local-first substrate creates room for larger operating models. The following are proposed and have no committed dates.
Shared knowledge
Federation between local work and shared team baselines.
Portfolio context
Cross-repository views and analysis across a governed collection.
Enterprise controls
Hosted operation, access controls, and managed topology after validation.
Evaluate the boundary
Bring a repository. Inspect what CoreStory can support.
We’ll separate what runs now from what still needs proving.
Talk to an expert →