Code search, blast radius and data model

nima keeps its own index of the repositories your workspace connected through GitHub. Three routes read it with an API key or OAuth token, all with scope knowledge:read: one finds files, one says what a change touches, and one returns the tables and columns the code declares. None of them returns source code. Read the code from your own checkout, using the paths and line numbers these routes give you. How the index is built is explained in Code intelligence.

Code search, blast radius and data model routes

The three code intelligence routes, their scope and the MCP tool that calls each one:

Paths are relative to https://api.getnima.io. All three are on the soft rate limit tier, 300 requests per 10 seconds. See Rate limits and Scopes and permissions.

Search the codebase

The search response fields

Each entry in results has these fields.

match only carries meaning when the query looks like a path (it has a / or ends in a file extension) or a single identifier (one token, optionally with ()). For those, exact means the hit's path equals the path you gave, or the file declares a symbol with that name. semantic means a path or symbol channel matched but the hit is not the literal. none means no channel confirmed the literal: the hit is only a neighbour, and the thing you named may not exist. Any other query, such as a plain-English question, is always semantic.

symbols[].kind is one of function, class, interface, type, enum, route, entity or decorator. symbols is the file's own list, not the symbols that matched, and a file declaring more than eight shows an arbitrary eight.

Get the blast radius of a change

Given files, symbols or both, return what a change there reaches: the modules that depend on it, the dependent files by distance, the tests that cover it, and the spec documents that govern it. It is read from the existing index and the knowledge graph's code links, with no re-index and no model call, so an index behind your commit answers about the commit it was built from.

Pass a pull request's changed files as paths to review its reach, or a symbol name to answer "who depends on this" before you edit.

At least one of paths and symbol is required, and together they may name at most 50 targets. Duplicates and blank entries are dropped before counting.

The blast radius response

Errors from the blast radius route

curl -s -G https://api.getnima.io/v1/code-intel/blast-radius \
  -H "Authorization: Bearer $NIMA_API_KEY" \
  --data-urlencode "paths=src/upload/upload.guard.ts,src/upload/limits.ts" \
  -d "depth=2"

Read the data model

Return the tables, collections, columns, relations, indexes and enums that a repository's ORM definitions and SQL declare, each with the file and line it is declared at. The model is parsed at index time with no model call, so it is deterministic for a given commit.

Fourteen ORMs are parsed: Prisma, Drizzle, TypeORM, Sequelize, MikroORM and Mongoose for TypeScript and JavaScript; SQLAlchemy, SQLModel, Django, Peewee and Tortoise for Python; GORM, ent and Bun for Go. SQL CREATE TABLE and ALTER TABLE statements in migrations and schema files are parsed too, and an ORM table shadows a SQL table of the same name. An empty model means no supported ORM or DDL was found, not that the app stores no data.

The data model response in summary mode

Without table, the response is { table: null, repositories: [...] } with one entry per repository. Each entry has mode: "summary", the repository fields below, and tables, capped at 400 per repository with truncated set when more exist. A summary table carries name, schema, entityName (the declaring class, null for SQL), orm, kind (table, view, collection or embedded), filePath, lineStart, columnCount and relationTargets.

The data model response in detail mode

With table, each repository entry has mode: "detail" and returns the matching tables in full plus the matching enums. matchedBy is exact when a table's name, schema.name or declaring class equals your query, substring when none did and every table whose name or class contains it is returned, and none when nothing matched. totalMatches is the count before the cap. Detail mode returns at most 20 tables and 50 enums per repository, with truncated set when either cap was hit.

A full table has name, nameSource (explicit or convention), schema, entityName, orm, kind, filePath, lineStart, lineEnd, primaryKey, columns, indexes and relations.

Each detail entry also has enums: name, values, orm, filePath, line, for enums whose name contains your query or that a returned table's enum columns reference.

The repository fields on every data model entry

Every entry, in either mode, starts with repositoryId, repoFullName, commitSha, extractedAt, orms, tableCount, enumCount and warningCount. commitSha is the commit the model was extracted at: compare it to your own before you trust it. extractedAt: null means no model has been extracted for that repository yet, and its counts are zero.

Errors from the data model route

curl -s -G https://api.getnima.io/v1/code-intel/data-model \
  -H "Authorization: Bearer $NIMA_API_KEY" \
  --data-urlencode "table=invoices"

Retrieval channels

The matchedChannels field of a search hit says why a file surfaced. Your query fans out across several channels at once, the ranks are fused, and a model reranks the candidates against the question. A strong hit usually carries several channels.

What the code index holds

The code index holds one row per file for 22 file extensions of source code, plus a fixed set of documentation files. Nothing else in a repository is searchable.

A file type outside that list, such as .rb or .kt, can never appear in results, however well it matches. Summaries are written when the file is indexed, from the first 300 lines of a code file or the first 2,000 lines of a documentation file, so a longer file has a summary describing only its opening, and a file changed since the indexed commit has a summary of the old version. Only the default branch is indexed. Read Code intelligence for how indexing runs.

Code routes not reachable with a key

Five code routes exist for a signed-in browser session and refuse every API key and OAuth token. A route that declares no scope answers 403 with This endpoint does not accept OAuth access tokens, whichever scopes you hold, and the admin panel has no control for any of them.

The repositories array on a search response is the only way to read index state with a key: it carries the id, status, indexed commit, last sync time and warnings for every repository. To refresh an index, push to the default branch and let the index follow, or ask the nima team. There is no route for reading a file's contents.