01 / Primitive
One bounded behavior
A validator, parser, retry policy, codec, or redaction function.
Compose into an agent-built systemFunctionFoundry compares regeneration with verified software acquisition before an agent spends tokens, tool calls, CI minutes, and human review recreating proven behavior.
The ranges overlap, so evaluation evidence should resolve functional and integration uncertainty before spending.
A package name is not a functional specification, and “just generate it” is not a cost model. FunctionFoundry makes functionality, build effort, evidence, rights, and integration cost queryable before code enters a repository.
Turn purpose, runtime, deployment, security, rights, and budget into an inspectable requirement set.
Model token use, agent runtime, CI cycles, review, maintenance, and uncertainty as ranges—not a magic number.
Rank primitives, modules, services, and systems by coverage, evidence, rights, acquisition cost, and integration work.
Proceed with a build, acquire one capability, or combine smaller pieces under explicit policy and budget limits.
FunctionFoundry does not split snippets from software. It indexes one capability graph across four levels, so an agent can compose verified pieces or acquire a complete service using the same contract, evidence, rights, and provenance model.
A validator, parser, retry policy, codec, or redaction function.
Compose into an agent-built systemAuthorization middleware, an OAuth adapter pack, or an address-normalization module.
Integrate behind a typed interfaceA document extraction service, malware gateway, or webhook worker.
Run hosted or deploy a pinned artifactA billing engine, audit control plane, or domain-specific application core.
Acquire a complete behavior boundaryGranularity changes the economics—not the quality bar. Even the smallest unit needs a contract, tests, versioning, license rights, and an immutable artifact record. A loose code snippet does not qualify.
A capability may be a composable piece or a complete service. These previews use canonical marketplace manifests, and each fit score is calculated against the homepage demo request: “I need a TypeScript service that extracts transactions from PDF bank statements, redacts PII, deploys to AWS Lambda, and includes commercial modification rights.”
Extracts normalized transactions from bank-statement PDFs with optional PII redaction and reconciliation metadata.
Hosted API execution
$0.20 / execution
pay per execution
Converts vendor invoices and scanned PDFs into validated invoice, line-item, tax, and payment-term records.
Hosted API execution
$0.20 / execution
pay per execution
Detects and redacts sensitive identifiers across text, JSON, PDF, and document images with explainable findings.
Hosted API execution
$0.20 / execution
pay per execution
Every version is an immutable evidence bundle: contract, test report, security posture, SBOM, digest, provenance, and commercial rights.
Machine-readable input and output contracts, coverage, fixtures, benchmarks, and private acceptance-test simulations.
Findings, dependency inventory, permissions, data classifications, and deployment boundaries sit beside the price.
Versioned artifacts arrive with SHA-256 digests, build metadata, receipts, licenses, and entitlement checks.
API keys are organization-scoped. Budgets, per-transaction limits, daily limits, idempotency, and explicit confirmation are enforced server-side.
Natural language plus hard compatibility constraints.
Wallet and agent policies bound every purchase.
Hosted, artifact, and source access are never conflated.
Files, dependencies, tests, security, and rollback.
A credible comparison includes model tokens, wall-clock agent time, tool calls, CI minutes, human review, security work, and ongoing maintenance. FunctionFoundry keeps estimates as ranges with visible assumptions.
Compare a modeled build range with the expected integration range.
Count likely retries, test runs, and repository changes—not only model output.
Keep security, maintenance, and code-review effort inside the decision.
A capability purchase should beat regeneration on total cost, evidence, rights, and time—not merely appear above it in search. The right outcome may still be to build.
| Decision factor | Rebuild from scratch | Verified capability |
|---|---|---|
| Agent compute | Tokens, retries, and tool calls | Acquisition plus integration model |
| Elapsed work | Implementation, CI, and review | Seller estimate plus buyer validation |
| Functional proof | Create and maintain tests | Versioned test evidence |
| Security posture | Establish after implementation | Findings and SBOM up front |
| Commercial rights | Newly built and internally owned | Explicit source or artifact rights |
Agents can query verified supply before implementation, then search, evaluate, purchase, license, and integrate behind organization scope, idempotency, explicit confirmation, and budget policy.
Search is read-only and can run before every substantial build. The MCP server calls the REST API instead of duplicating business logic, and purchase tools require confirmPurchase: true.
POST /api/agent/search
Authorization: Bearer ff_••••••••
{
"request": "Extract transactions from PDF bank statements",
"constraints": {
"requiredRuntimes": ["node"],
"requiredRights": ["source-access", "modification"],
"maximumCriticalFindings": 0
}
}
→ 100 / 100 fit Bank Statement Transaction NormalizerAsk what the build would cost, compare a verified alternative, run the guided evaluation, spend demo funds, receive a license, and inspect the simulated integration pull request.