Repository intelligence for AI agents

Give your agents a map before they touch the code.

LEIO Code turns any repository into structured, evidence-backed context for ChatGPT and OpenAI SDK agents. Find ownership, trace runtime wiring, follow call graphs, and catch architectural drift before you ship.

Read-only analysis Source-backed answers Public Git or private GitHub

LEIO / INVESTIGATION CONTEXT PREPARED

“Trace the authentication path and tell me what could break if we change token validation.”

12relevant symbols

7runtime relationships

3configuration contracts

4targeted tests

2architectural risks

SYMBOLvalidate_token()mycelia-api/auth/tokens.py:184

CONFIGAUTH_JWT_ALGORITHMdeploy/profiles/production.env:42

RELATIONcalled by → require_current_user()mycelia-api/auth/dependencies.py:76

RESULT

Changing token validation affects two authentication entry points, one production environment contract, and four targeted regressions.

Why repository intelligence

Search finds text. LEIO finds the system around it.

Ordinary repository searchLEIO Code

Matching stringsSymbols and semantic identities

Isolated filesCall and import relationships

Configuration namesRuntime and deployment lineage

Generic repository summariesCapability-aware workspace profiles

Large file dumpsRanked task-specific context

Static lint rulesArchitectural drift doctors

Your agent receives the smallest useful context, with the evidence needed to defend it.

How it works

From question to repository evidence in four steps.

01 / SELECT

Point LEIO at the right repository and revision.

Use a public Git URL or connect GitHub for private repositories. The selected target remains scoped to the current session.

02 / ROUTE

LEIO chooses the correct intelligence surface.

A symbol lookup, call-chain question, deployment explanation, and architectural audit are different operations. LEIO routes them accordingly.

03 / RESOLVE

The warm index connects code, configuration, topology, and contracts.

LEIO retrieves only the symbols, files, relationships, tests, and risk notes relevant to the task.

04 / GROUND

Every answer returns repository evidence.

Results include source paths, line locations, entity types, relationships, and diagnostic provenance.

Capability system

One repository. Five ways to understand it.

Find

Locate the real owner of a concept.

Search symbols, environment variables, API routes, Redis keys, Docker services, deploy targets, and domain cartridges.

EXAMPLE PROMPT“Where is PAYMENT_APPROVED produced and consumed?”

search_repository

Repository GPS

LEIO knows what kind of repository it entered.

Different repositories expose different risks and capabilities. LEIO identifies the active workspace profile and adapts its search kinds, graph operations, deployment knowledge, domain facets, and architectural doctors to the repository in front of it.

Capability-aware intelligence prevents agents from invoking generic workflows against repositories with specific operational rules.

WORKSPACE PROFILEmycelia

AVAILABLE FACETS

24 deploy targets
24 runtime profiles
24 domain cartridges
120 architectural doctors

INDEX

100,367 symbols
5,024 environment references
316 Redis keys

Illustrative workspace snapshot — capabilities vary by selected repository.

Context preparation

Less context. Better context.

Large context windows do not solve poor repository selection. LEIO prepares a focused evidence bundle for the exact task, reducing irrelevant files while preserving the dependencies and constraints that could change the result.

TASKAdd retry behavior to outbound payment confirmation

PRIMARY FILES

payments/confirmation.py
payments/retry_policy.py

DEPENDENCIES

PaymentStatus
payment:confirmation:{tenant_id}
PAYMENT_RETRY_LIMIT

TARGETED TESTS

test_confirmation_retry.py
test_payment_idempotency.py

RISKS

Duplicate external effects
Tenant-key isolation
Retry budget drift

Context is ranked by relevance, responsibility, and risk, not file popularity.

Architectural doctors

Turn repository rules into checks agents can run.

LEIO doctors inspect contracts that ordinary linters cannot see: cross-service configuration, deployment bundles, generated artifacts, domain boundaries, runtime topology, and repository-specific invariants.

When a production drift is discovered, encode it as a doctor or canonical evidence path. The next agent should detect it before repeating it.

AUTH SESSION CONTINUITYWARN

EVIDENCE

deploy/profiles/api.env:28
deploy/profiles/gateway.env:31

RISK

Existing sessions may fail after deployment.

NEXT ACTION

Align issuer configuration before publishing.

OpenAI integration

Repository intelligence inside the conversation.

LEIO Code exposes structured tools through MCP and an interactive OpenAI Apps SDK interface. Agents can select a repository, inspect its capabilities, ask structural questions, and open the supporting evidence without leaving the conversation.

01Inspect repository status

02Prepare task context

03Search symbols and contracts

04Explain runtime lineage

05Traverse call and import graphs

06Run architectural audits

07Select or clear the active repository

CHATGPT + LEIO CODE

Open the repository intelligence layer where your agents already work.

Open LEIO in ChatGPT ↗View MCP tool reference ↗

Designed for production repositories

Useful to agents. Legible to humans. Constrained by default.

Read-only analysis

Repository intelligence tools inspect the selected checkout without editing source files.

Evidence-backed output

Answers carry paths, symbols, line references, relationships, and diagnostic evidence.

Session-scoped targeting

Repository selection is isolated to the active MCP session and can be cleared.

Private repository support

Connect GitHub for authorized private repository access.

Minimal secret exposure

LEIO may identify configuration names and contract drift. It does not retrieve live production secret values as evidence.

Repository-aware controls

Workspace profiles and doctors preserve local architectural rules instead of applying generic assumptions.

Use cases

Built for the moments when guessing is expensive.

Before an edit

Identify owners, dependencies, tests, and constraints before changing code.

During code review

Trace the behavioral impact beyond the visible diff.

Before deployment

Audit configuration, artifacts, profiles, and runtime contracts.

During incidents

Resolve routes, environment variables, services, and call chains quickly.

During onboarding

Replace repository tours with evidence-backed questions.

Across agent platforms

Provide a consistent repository-intelligence layer to ChatGPT, Codex, and MCP-compatible agents.

Repository intelligence for every agent task

Your agent is ready to code. Is it ready for your codebase?

Connect a repository and give every investigation a current map, explicit constraints, and source-backed evidence.

Bring LEIO to your engineering platform ↗