Your code ships.
Your metrics don't lie.
Horion finds the gap.
Connect your repo. Horion analyzes metrics, logs, and traces instrumentation in your code, scores each pillar, and opens a PR with the actual fix.
100k tokens/month free · no card · ship in 5 minutes
Connect it once.
Fix it forever.
Wire your repo and your telemetry. Horion runs the analysis, scores every pillar, and hands you findings with the evidence, the cost, and the diff — then closes them through your coding agent or through a pull request of its own.
scroll to follow the loop ↓
Connect your repo and telemetry
Point Horion at a repository and at your observability vendor. It reads the code and the live signals side by side, and scores the repo before you touch anything.
read-only keys · scoped per project
Watch the analysis run
Nine stages, streaming live: clone, AST, telemetry join, coverage, efficiency, score. A knowledge base and a bench of language experts back every call.
One agent, many specialists — the knowledge base and a language expert per stack, called in as the analysis needs them.
Findings with evidence, cost, and a fix
Every finding lands with the file and line, the model's reasoning, what it costs you per month, and the diff that closes it.
Database client used without auto-instrumentation
“Every query on this engine leaves the trace early, so checkout latency shows a hole exactly where the database work should be. One instrumentor call puts the span back.”
engine = create_engine(DATABASE_URL)session = Session(engine)
+ SQLAlchemyInstrumentor().instrument(engine=engine)engine = create_engine(DATABASE_URL)session = Session(engine)
Fix it your way
Two ways out. Hand the work to the coding agent you already use, or let Horion open the pull request itself.
Your coding agent, guided by Horion
Install the Horion Skill and connect the MCP server. The Skill hands your agent the harness — it pulls the ruleset, guides the instrumentation, and walks each finding to a fix. Your code never leaves the machine.
Claude Code
Cursor
Devin
Codex
CopilotOr let Horion open the PR
Pick a finding and queue the fix. Horion patches it, has its experts review the diff, and opens a pull request you read like any other.
Horion applied the observability fixes and opened a pull request on acme/payments-api.
The agent path and the fix-PR path land the same score — take whichever fits your team.
live · the gate re-evaluates on every push ↑
Lens reviews the PR.
The gate decides if it ships.
The moment a pull request opens, Horion Lens posts an observability review — per-pillar scores, the gaps, the fix. Set a minimum score and that review becomes a required check: below the bar, the merge button stays locked.
Posts the review
A single comment on the PR — Metrics, Logs and Traces scored against the diff, each gap traced to a file and line.
Lands a commit status
horion-lens/observability shows up as a check: pending while it runs, then pass or fail against your repo's minimum score.
Holds the merge
Mark the check required in branch protection. A PR under the bar can't merge until a new commit clears it — observability stops regressing silently.
Articles & hands-on
from the engine room.
Long reads, postmortems, and step-by-step labs from the team building Horion — how we score, instrument, and review production systems.
Engineering the score: how Horion weighs metrics, logs, and traces.
The first score we shipped, nobody trusted. Here is how we rebuilt the rubric — pillar by pillar — and what changed in PR reviews.
Instrument a FastAPI service with OpenTelemetry in 15 minutes.
Wire the OTel SDK, point it at the collector, and verify your first distributed trace — copy/paste guided, end to end.
The four hours we spent chasing a missing log line.
One async consumer, no structured logs, a queue silently filling up — and the logging pattern we now require for every async job.
See your stack score itself.
100k free tokens every month. No credit card. First analysis in under five minutes.