Public beta · v0.1.4-beta

Context intelligence for coding agents.

CORA is a local-first MCP context engine that helps coding agents understand your application before they edit it. It builds deterministic codebase context, adds optional on-device AI enrichment, and exposes MCP tools for safer patch planning, impact analysis, and architecture-aware development.

Local-first MCP-native Signed releases Registration-gated distribution

Coding agents need application context, not just file snippets.

Modern coding agents are powerful, but they often operate with incomplete awareness of system boundaries, dependencies, tests, APIs, and architecture decisions. CORA gives agents a reliable context layer before they propose or apply changes.

⌁

Understand impact

Map files, symbols, dependencies, routes, documentation claims, and likely blast radius before code changes are made.

◇

Plan safer patches

Expose MCP tools that help agents build patch plans grounded in actual repository structure and deterministic evidence.

◎

Reduce blind edits

Use context-first workflows and optional enforcement hooks so agents query CORA before editing sensitive code paths.

How CORA works

CORA runs locally as a split-container MCP runtime. The core container handles deterministic indexing, compiled context, and MCP tools. The enrich container can add optional on-device AI summaries and inferred relationships without overwriting deterministic facts. If your repo has specs or roadmaps, CORA can also surface a product scope beacon so agents see what is in scope before they edit.

Local repositoryRead-only mount

→

CORA CoreDeterministic graph + MCP tools

→

CORA EnrichOptional on-device AI

What’s new in v0.1.4-beta

This beta makes local enrichment real AI — and helps agents stay aligned with your documented product intent.

Local AI enrichment

Optional AI summaries and relationships run on your machine after indexing. Indexing finishes first; enrichment can continue in the background so you are not blocked waiting.

Product scope beacon

CORA reads specs and roadmaps into a simple north star: what is in scope, what is out of scope, and what “done” looks like.

Clearer change guidance

cora_change_brief returns a structured narrative with evidence. Change briefs and patch plans can gently flag work that drifts from the documented plan — advisory only, not a lock.

Faster enrichment model

Updated on-device model for quicker enrichment on typical projects. The first enrichment download is still large because it includes the local model.

MCP tools for safer agentic coding

CORA exposes focused MCP tools that help coding agents inspect the application before planning or applying edits.

cora_search

Search indexed repository context across files, symbols, dependencies, and documentation.

cora_application_context

Return application overview, indexing status, and the product scope beacon when docs are present.

cora_file_context

Explain how a specific file fits into the broader application context.

cora_change_impact

Estimate likely change impact across related files, tests, dependencies, and documented contracts.

cora_patch_plan

Generate a context-first patch plan grounded in deterministic evidence and known boundaries.

cora_change_brief

Return a natural-language change brief with evidence, validation gates, and advisory scope alignment.

Observed gains so far

Early dogfooding and beta rollout work has shown concrete improvements in setup reliability, security posture, and agent workflow consistency.

↗

Faster setup across agents

Unified MCP wiring for Cursor, Claude Code, and Codex reduces configuration drift and shortens time-to-first-productive session.

🛡

More reliable security gates

Hardened update path and resilient dependency scanning make CI outcomes more trustworthy while preserving strict vulnerability checks.

⚡

Better agent consistency

Context-first skills and hook workflows reduce blind edits and improve patch planning quality before code changes are applied.

Designed for local-first security.

CORA beta is built for controlled early access. The installer is registration-gated, runtime images are private, releases are signed, and the MCP runtime requires activation before it starts. Activation validates entitlement only; it does not transmit source code, repository contents, prompts, MCP queries, file names, repository paths, or indexed context.

Read-only repository mounts
Non-root container runtime
Secret-aware indexing exclusions
Digest-pinned images
Signed release artifacts
Private GHCR distribution
Daily activation lease
No code or prompt telemetry

Controlled registration-based distribution.

Current public beta is v0.1.4-beta. After registration, testers download the installer without GitHub CLI or GitHub repo access, install locally, and activate the MCP runtime. See what’s new in this version.

Register
Create a Horizon Labs account and complete beta registration.

Download
Access the gated installer from your registration-linked download page.

Install
Run the installer on macOS, Linux, or Windows with Docker available.

Activate
CORA validates entitlement before starting the MCP runtime.

Where CORA helps

CORA is designed for teams and builders using coding agents in real repositories where architecture, security, and change control matter.

Agentic development

Give Cursor, Claude Code, Codex, and other MCP-compatible agents a safer context layer.

Security-aware patching

Help agents reason about blast radius before touching authentication, APIs, infrastructure, or sensitive workflows.

Codebase onboarding

Quickly understand unfamiliar repositories through searchable, graph-backed context.