THE PRODUCT
LOCI | Line-of-Code Intelligence - runs at plan, at PR, and at merge. Binary-level signals. Human decides what ships.
Works with Claude Code, Cursor, Copilot, GitHub PRs, CI/CD.
Not a GPT wrapper. A vertical model built for binaries.
HOW YOU WORK WITH IT
LOCI runs as a skill inside Claude Code. You write code, LOCI analyzes the compiled binary and surfaces execution signals in real time.
Preflight audits the plan. Post-edit reviews the PR. Gate gives the human the final call.
When the AI agent plans a change, LOCI models the resulting binary before a line of code is written. Stack depth, timing, power, regressions, predicted from the plan.
Fix the plan, not the PR.
Agent writes code and opens a PR. LOCI diffs the new binary against baseline and surfaces regressionsת before merge, at the binary level.
Every AI PR, reviewed at the binary.
LOCI presents the evidence. You approve or block. Over time, LOCI calibrates to your quality bar and surfaces only what matters.
Evidence on every PR. You control what ships.
WHAT LOCI MEASURES
LCLM reads the binary and predicts five classes of execution behavior. Each can be set as a budget, LOCI enforces it on every change.
Critical-path execution time, p95/p99 latency, tail amplification.
+3.4ms in AES_encrypt() on TLS handshake path
Requests/sec, tokens/sec, instructions per cycle. Regressions against baseline.
BLE throughput −8% in ble_ll_tx()
Worst-case stack usage per function. Enforced against MISRA / AUTOSAR budgets. No instrumentation needed.
motor_ctrl() exceeds 2KB stack budget
Power & Energy
Execution cost, energy per operation, expensive paths.
+12% power on sensor wake path
Worst-case execution paths, rare branches, attacker-reachable inputs.
Input-dependent branch skips bounds check in parse_packet()
Execution Gates
LOCI lifts two binaries to IR, builds an execution graph, and fires or clears each gate from the diff, before a single instruction runs. No execution required. No instrumentation. Just two ELF files.
CUSTOMER PORTAL
Track regressions caught, debugging hours saved, and team-wide signal trends. Manage seats, plans, and quality budgets from a single dashboard.
/plugin install loci@claude-code-marketplace
Adds /loci:preflight and /loci:post-edit slash commands. Stop gate hook runs on every PR. Human approves in the terminal.
npx skills add auroralabs-loci/loci-claude
- uses: auroralabs/loci-action@v1 with: gate: post-edit
Runs on every PR. Blocks merge if signals exceed baseline. Posts findings as PR comment.
Subscribe via AWS Marketplace
SaaS subscription through your AWS account. Billing consolidated on your AWS invoice. Cloud hosted, AMI/VM in customer VPC, or SageMaker integration.
LCLM | LOCI’s Large Code Language Model – is trained on billions of ASM blocks and real-time traces from open-source projects spanning IoT, networking, safety-critical, and industrial systems. It learns the execution behavior of compiled code, not just the syntax of source code.
That’s how LOCI models the binary before the code is written. And why LOCI’s signals are inspectable, explainable, and verifiable, every time.