Platform
The civic data layer for agents, starting with New York City.
Akil sits above official public-data sources and turns fragmented civic records into structured context that people, organizations, developers, and AI agents can use without rebuilding the data plumbing.
Semantic Layer
One layer between official data and every product built on it.
Raw rows are only the beginning. Akil interprets what records mean in their source context, connects them through stable civic anchors, and keeps the limits of each record visible.
Official public data
NYC Open Data, agency portals, spatial layers, state records, federal records, public meeting records, licenses, payments, permits, enforcement actions, and source documents.
Akil semantic layer
Normalized schemas, source IDs, BBL/BIN/address joins, entity-role discipline, codebook interpretation, metric caveats, regulatory interpretation, mirror coverage, freshness, and workflow logic.
Applications and agent workflows
Property diligence, business site screening, funding research, district briefings, procurement intelligence, housing conditions, public-accountability checks, MCP tools, and white-labeled workflows.
Capabilities
What makes the layer useful.
Source-aware meaning
Akil preserves official source labels while explaining status fields, code systems, record families, dates, counts, rates, and regulatory signals in plain English.
Stable civic anchors
Addresses, BBLs, BINs, districts, agencies, organizations, vendors, licenses, applications, contracts, and source record IDs remain explicit so outputs can be checked and reused.
Trust contracts
Responses carry source attribution, applied windows, mirror coverage, freshness notes, candidate-match caveats, and the boundary between research support and official determinations.
Trust Checks
The semantic layer is tested for public-record behavior.
Akil's trust checks look for public-record mistakes that agents can easily make: treating a filing as clearance, a name match as identity, a capped page as the full universe, or an empty result as proof that nothing exists. The point is not to claim perfect data. The point is to make record boundaries visible and repeatable.
Sources
Source and caveat checks
Checks for source attribution, coverage disclosure, freshness notes, result caps, caveats, and verification paths.
Limits
Coverage and no-result checks
Checks that broad, capped, empty, or degraded responses stay bounded to the searched source, filters, and window.
Status
Public trust status
The public trust-status snapshot lives next to the MCP Worker: mcp.askakil.ai/trust/status.
Proof Points