Search your code the way you think.
Semantic search across every repo, branch, and language — with grounded answers, call-graph traversal, and architecture diffs. Indexed in real-time on every push, scoped by your IdP, and queryable from CLI, REST, GraphQL, or your IDE.
Indexes the repos and languages your team already ships
Language coverage
First-class parsing for every language your team writes
Tree-sitter grammars + LSP semantics + embeddings. Search by symbol, by reference, or by meaning — across all of them at once.
Plus Scala, Elixir, Haskell, OCaml, Zig, Lua, Bash, SQL, HCL, YAML, Dockerfile, and Markdown.
Ask your codebase
Grounded answers — with citations to real files and line numbers
Ask in plain English. Answers are grounded against your indexed repos with file + line citations, so you can verify (and your reviewers can audit) every claim. Pick the model — Opus for depth, Haiku for speed.
- Citations link straight to the file in GitHub, GitLab, or Bitbucket
- Cross-repo answers — no need to know which repo to ask about
- Branch- and PR-scoped chats for code review and architecture diffs
- Pluggable models: Opus, Sonnet, Haiku, GPT-class, self-hosted
Python · curl · GraphQL
Search, ask, traverse — with the surface that fits the task
Use the SDK from a service or an agent, REST from a script, GraphQL when you want to walk the code graph (call sites, type usages, tests). Same index, same semantics, three surfaces.
- Real-time indexing — every push reflected in under 30 seconds
- GraphQL traversal: callers, callees, type usages, test coverage
- Webhook fanout for "new symbol" or "removed symbol" events
- IdP-scoped access — never index repos a user can't see
1# pip install vxcloud-icodebase2from vxcloud.icodebase import Client3 4ic = Client() # auto-discovers VX_API_KEY5 6# 1) Semantic search across every connected repo.7hits = ic.search(8 query="how do we verify a JWT across services?",9 repos=["studio-backend", "vxcli", "platform-rs", "vxcloud-web"],10 languages=["python", "go", "rust", "typescript"],11 top_k=10,12)13for h in hits:14 print(h.score, h.repo, h.path, h.symbol)15 16# 2) Ground a question in the codebase ("RAG over code").17answer = ic.ask(18 question="Where do we verify JWTs?",19 grounded=True,20 cite_files=True,21 model="claude-opus-4-7",22)23print(answer.text)24for c in answer.citations:25 print(c.repo, c.path, c.line)26 27# 3) Find call sites of a symbol — including indirect ones.28sites = ic.references(symbol="authenticate_request", repo="studio-backend")29print(f"{len(sites)} call sites · {sites[0].kind} from {sites[0].caller}")30 31# 4) Diff the architecture across two branches.32report = ic.diff_architecture(repo="studio-backend", base="main", head="feature/agents")33print(report.summary()) # added modules, removed deps, drift signalsDon’t know which repo? Just ask.
ic.ask(question=“where do we rate-limit incoming webhooks?”) — it searches every repo and cites the file.
Built for org-scale code
Code intelligence that scales with your monorepo
Semantic + symbolic search
Combines embeddings with tree-sitter / LSP for meaning-aware results that still respect identifier boundaries.
Call-graph traversal
Walk callers, callees, type usages, and test coverage across repos — including indirect call sites.
Architecture diffs
Diff modules, deps, and entry points between branches, tags, or PRs to spot architectural drift early.
Grounded code Q&A
Plain-English answers with file + line citations, model of your choice, IdP-scoped to your access.
Real-time indexing
Pushes show up in search within thirty seconds. No batch jobs, no overnight rebuild.
IdP-scoped access
Repo and namespace ACLs honored end-to-end. A user only ever sees code they can already pull.
Cross-repo references
Find call sites for a symbol across hundreds of repos — same query, one result list, ranked.
Event webhooks
Fire on new / removed / renamed symbols and architectural changes — feed into your dev workflow.
Test + coverage signals
See which symbols have tests, what framework they use, and the coverage delta over the last 30 days.
Real numbers from real index loads
Built for monorepos — measured on them
Aggregated across customer indexes ranging from 100k to 250M lines of code, on instance profiles tuned for embedding throughput.
Where iCodebase pays off
From onboarding to incident response
Onboarding new engineers
Day-one search and Q&A across the entire org's code — new hires self-serve answers without interrupting senior engineers.
Refactor planning
Find every call site (direct and indirect) before you rename, split, or remove a public API. Plan blast radius, not guess it.
Code review acceleration
Reviewers get architecture diffs and "what changed" summaries on every PR — focus on intent, not noise.
Powering coding agents
Ground every Claude / GPT call against your real code. No more hallucinated APIs, no more outdated docs.
Tech-debt visibility
Surface deprecated symbol usage, untested modules, and architectural drift — with PRs auto-opened on the worst hotspots.
Audit + compliance trails
Prove who could read which code at which time — IdP-scoped access logs for compliance and contractor reviews.
Make your codebase searchable the way you think.
Connect your GitHub, GitLab, or Bitbucket org — index every repo your team can read, search by meaning, and ground every coding-agent call against real code.