HorusUse case
Give AI coding agents grounded project context
Horus gives AI coding agents grounded context across source code, logs, queues, metrics, deploys, ownership, and evidence — so they can investigate production issues without rebuilding the system model every session.
AI coding agents are good at writing code and bad at remembering your system. Every session they rediscover structure, re-read files, and guess at runtime behavior. Horus gives them a persistent, evidence-backed model to start from.
The problem: agents start every session cold
A fresh agent knows nothing about your last deploy, your queue topology, or which service owns checkout. It rebuilds that picture from scratch — slowly, and often wrong.
Why raw repo access isn't enough
- The repo shows code, not runtime behavior — no logs, no metrics, no queue state.
- It doesn't say what changed recently or why a symptom started.
- Reading the whole repo to answer one question wastes the context window.
Why scattered MCP servers burn context
Connecting an agent to ten tools via MCP gives it access, not understanding. It still has to fetch from each, reconcile formats, and decide what matters — every session. More on this in Horus vs MCP.
What grounded project context means
- A persistent code graph and service/queue map.
- Runtime evidence linked to the code it implicates.
- A timeline of changes correlated with symptoms.
- Ownership of the affected symbols and services.
How Horus helps agents investigate production issues
Horus runs locally, indexes the repo, connects runtime sources, and produces a deterministic investigation an agent can read directly — instead of guessing.
Example flow: a “slow checkout” investigation
horus investigate "slow checkout"Horus correlates the symptom with recent changes, queue state, and the code path, then resolves to the most likely symbols and files.
What the agent gets
| Output | How the agent uses it |
|---|---|
| Timeline | Anchor the symptom to the change that preceded it |
| Evidence | Logs, metrics, queue and database state linked to code |
| Suspected changes | Recent commits and symbols most likely involved |
| Affected owners | Whose code to read first, or who to involve |
| Next checks | Concrete steps instead of open-ended exploration |
The core idea
AI coding agents are strongest when they don't need to rediscover the system from scratch. Horus gives them a structured investigation layer.