Source-derived context infrastructure
Noodlbox builds a context box from source: definitions, workflows, package surfaces, and taxonomy. It becomes the system of record for codebase structure and behavior across the tools that work on code.
curl -fsSL https://get.noodlbox.io | bashLLMs can query it. They do not build it.
Context box
Source in. Context box out. The artifact keeps repository context usable across tools.
Analyze
Read the repository itself and derive definitions, workflows, package surfaces, and taxonomy without asking a model to infer them.
Package
Package the output as a context box with stable structure that can move across tools.
Consume
Serve the same artifact to agents, APIs, UI surfaces, CI workflows, and docs instead of rebuilding context inside each workflow.
Repository
Source, history, structure
Analysis
Definitions, workflows, package surfaces
Context box
Structured artifact
Tools
MCP, API, UI, CI
Artifact metadata
context-box.nbx
The same build output can move from code search to MCP, API slices, review workflows, generated docs, and repository overview without each tool inventing its own version of the codebase.
System of record
Git records files and diffs. Tools still need structure, behavior, and context that stays attached to the code.
Prompts, stale docs, and text-first indexes are not a durable system of record for how a codebase is structured or how it behaves. Noodlbox fills that gap by deriving context from source and making it usable across tools.
The result becomes the system of record for codebase structure and behavior across the tools that work on your code.
Temporary, task-specific, and thrown away after the interaction ends.
Useful when they exist, but they drift, rot, and rarely stay aligned with the repository.
Fast for lookup, but they flatten structure and behavior into search results instead of durable artifacts.
Derived from source and reusable across the tools that work on code.
Derived from source
Not a blob of prompt context. A structured artifact with definitions, workflows, package surfaces, taxonomy, and snapshots.
Definitions
Resolve code structure across files and modules, not just text matches. Definitions stay attached to references, relationships, and their place in the repository.
AuthService.validateSession()
SessionRepository.findByToken()
UserService.loadPermissions()
file: auth/service.ts
Calls → SessionRepository
Imports → token-cache.ts
Referenced by → LoginHandler
Member of → Auth & Session
Workflows
Capture how code behaves across function boundaries, branches, and module transitions so tools can work with behavior, not just isolated definitions.
handleCheckout()
→ validateCart()
→ quoteInventory()
→ chargePayment()
✓ persistOrder()
✗ enqueueRecovery()
Entry point · branch path · downstream effect
Package surfaces
Resolve external APIs at the version actually used by the repository, so tools can reason about package boundaries with evidence instead of assumptions.
handleCheckout()→npm:stripe@4.0→charges.create()
fetchUser()→@prisma/client@5.0→prisma.user.findUnique()
Version-specific package surface, not a guessed dependency edge
Taxonomy
Group the repository into navigable modules and higher-level structure so humans and tools can orient around the same architecture view.
Snapshots
Tie every derived output back to the repository state that produced it. Context becomes versioned, inspectable, and shareable by commit.
checksum
blake3-91d3e6c8...
Consumers
One artifact for agents, APIs, CI workflows, docs, and UI surfaces.
MCP protocol. REST API. Exportable artifacts. The same context box can move across tools without being rebuilt as prompt glue.
mcp://query
Query the artifact through MCP and other integrations instead of relying on task-local prompt context.
ui://overview
Browse repository overviews, workflows, definitions, and snapshots without reconstructing structure by hand.
GET /api/context-slice
Retrieve exact slices of repository context programmatically for internal tools and product surfaces.
ci://impact
Use repository context and impact views in automation, review, and release flows.
docs://reference
Publish documentation that stays attached to definitions, workflows, and module evidence underneath it.
Registry
Pinned packages and public repositories can be published as reusable context artifacts.
Noodlbox is not just a local build step. Hub is the registry layer: a place to publish, reuse, and query context artifacts across repositories and packages.
That gives tools a stable way to reason about the code they depend on, not just the code they own.
Observed in task runs
Supporting evidence, not the category claim.
In a small SWE-bench Lite task set, the same tasks required fewer tool calls, less time, and lower spend when the agent could query a structured context layer instead of assembling context ad hoc.
10 tasks · same task set · same evaluation harness
Measure
Baseline
With Noodlbox
Change
Time to complete task set
36 min
23 min
-36%
Spend per task set
€17.70
€9.97
-44%
Tool calls
302
153
-49%
Time to complete task set
Baseline
36 min
With Noodlbox
23 min
Change
-36%
Spend per task set
Baseline
€17.70
With Noodlbox
€9.97
Change
-44%
Tool calls
Baseline
302
With Noodlbox
153
Change
-49%
Coverage
Coverage by language and resolution layer
| Language | DefinitionsDefinitions & structure | Intra-file referencesScope-based resolution | Cross-file referencesImports & module resolution | Type resolutionTypes & method resolution | Cross-repo referencesMulti-repository resolution |
|---|---|---|---|---|---|
Python | |||||
TypeScript | |||||
JavaScript | |||||
Go | |||||
Java | |||||
Kotlin | |||||
Ruby |
