Grounded code intelligence

Give your AI the codebase, not fragments.

Index your codebase once. Ask real questions in Codex, Cursor, Claude Code, or any MCP-compatible tool and get answers grounded in current source, symbols, and relationships.

Start with CtxE How it works

CtxE — Real Context, Real Time, Real Understanding — **Figure 01.** One context layer for terminal and MCP workflows.

Source model: AST-aware
Retrieval: Hybrid
Relationships: Graph
Agent surface: MCP

Stable release

Install in one command

macOS, Linux, and Windows PowerShell.

curl -fsSL https://storage.tlelabs.com/ctxe/latest/install.sh | sh

Context system

Evidence before answers.

CtxE combines syntax-aware indexing, local search, semantic retrieval, and graph traversal so agents can inspect the right implementation before they act.

01 / PARSE

Code-aware chunks

Tree-sitter and composite language profiles preserve functions, types, modules, and their neighboring context.

symbols → chunks → edges

02 / RETRIEVE

Hybrid search

Semantic vectors and indexed full-text search work together, with reranking available when the question needs it.

vector + FTS5 + rerank

03 / RELATE

Knowledge graph

Definitions, usages, calls, implementations, imports, paths, and impact remain connected across files.

direct → multi-hop → impact

04 / ASK

Goal-driven evidence

Ask plans bounded retrieval rounds, promotes source evidence, and reports degraded or partial results explicitly.

bind → retrieve → assure

05 / OPERATE

Local dashboard

Inspect workspace health, indexed chunks, research history, event logs, and configuration from one local surface.

ctxe dashboard

06 / CONNECT

MCP-native tools

Use the same indexed knowledge from Codex, Claude Code, Cursor, VS Code, and other stdio MCP clients.

ctxe mcp

Working surface

One index, two ways to work.

Ask from the terminal or let an MCP-capable agent gather evidence inside the editor. Both workflows resolve against the same code-aware index.

Command line

Direct research

Local index

$ ctxe index .

Indexed 1,284 files (4,872 chunks) in 41s

$ ctxe ask "How does user session refresh work after login?"

Comprehensive Session Refresh Analysis

1. Entry points
   POST /auth/refresh — src/api/auth/refresh.ts:31
   AuthMiddleware.refreshIfNeeded() — src/middleware/auth.ts:89
   BackgroundTokenRefresher — src/jobs/token-refresh.ts:22

2. Core logic
   SessionService.refresh() — src/auth/session.ts:147
   Validates against Redis + DB, rotates tokens, and blacklists the previous token.

3. Relationships
   Call graph depth: 4 levels | 17 direct files | 41 indirect files

4. Impact
   14 backend files + 3 jobs + mobile SDKs

Analysis complete. Ready for a detailed plan.

MCP client

Agent-assisted research

Connected

You

“Make the login flow more secure.”

Agent

Let me understand the implementation and its relationships first.

mcp / ctxe ask_context

“Where are all files and logic related to the current login flow?”

The login flow spans API handlers, JWT middleware, session services, and supporting infrastructure.

mcp / ctxe ask_context

“What are the security-sensitive parts and current weaknesses?”

Current implementation lacks token rotation and has weak rate limiting around refresh endpoints.

Agent

I now have the affected files, relationships, and security gaps needed for a precise plan.

Start locally

Make context a system, not a prompt.

Read the setup guide