agrt 0.1.0 — Production Ready

Autonomous Local Execution Platform

Deterministic orchestration. Replayable execution. Self-healing infrastructure.

Star on GitHub View Replay Engine
INTENTSCAFFOLDBOOTSTRAPPROVISIONPRE-FLIGHTJUDGEREPAIRREPLAYDELIVERY
Orchestrating
NODES: 0/9
Execution Lifecycle

Seven-stage autonomous loop.

Every execution traverses a governed DAG — from intent extraction to verified delivery. Each stage emits telemetry, validates state, and preserves replay evidence.

01

Intent

AI intent extraction & TaskSpec compilation from natural language.

02

Planning

Execution DAG construction with dependency resolution.

03

Sandbox

Isolated environment instantiation with cryptographic attestation.

04

Validation

Continuous behavioral validation & observability checks.

05

Repair

Formal failure taxonomy routing to specialized repairers.

06

Replay

Full execution trace preservation with sub-ms telemetry.

07

Delivery

High-confidence deployment with git-provider integration.

kernel_orchestrator_v6
[BOOT] Runtime initialized — PID 42891
[INTENT] TaskSpec compiled from user prompt
[SCAFFOLD] DAG constructed — 7 nodes, 11 edges
[BOOTSTRAP] Sandbox created at /tmp/agrt_7a3f
[PROVISION] Installing 12 dependencies...
[ATTEST] SHA256: a7f3c2e...9e1b ✓ VERIFIED
[JUDGE] Validation cycle 1 — running tests...
[JUDGE] 7/7 assertions passed
[REPLAY] Trace persisted — 847 events captured
[DELIVERY] git commit → "feat: auth service"
Confidence0%
AWAITING VALIDATION
Autonomous Recovery

Failure is deterministic.
Recovery is governed.

When orchestration fails, the kernel routes failures through a formal taxonomy to specialized repairers. No guessing — just engineering.

repair_hud / autonomous_recovery
repair in progress...● REPAIRING
Failure Detection
<200ms
Time to classify and route failure
Repair Routing
Taxonomy-Led
Formal classification drives repair strategy
Recovery Success
100%
All governed failures recovered autonomously
Governance Model
Watchdog-First
5s SIGTERM → SIGKILL escalation policy
Replay Engine

Every decision. Inspectable.

Forensic-grade execution replay. Scrub through every step, inspect stdout/stderr, trace confidence evolution, and audit every repair decision.

Time-Travel Scrubbing
stdout/stderr Capture
Confidence History
Node Inspection
replay_viewer / execution_trace
T+0.000sIntent Received
T+0.142sSandbox Created
T+1.893sDependencies Provisioned
T+2.104sEnvironment Attested
T+8.442sValidation Failed
T+8.671sRepair Executed
T+12.033sDelivery Complete
Operational Evidence

Infrastructure metrics. Not marketing.

Empirical data from production execution runs. Every number is measured, not claimed.

Repair Success Rate
0%

All taxonomy-classified failures resolved autonomously without human intervention.

Orchestration Latency
0ms

Median time from intent receipt to DAG construction and sandbox initialization.

Replay Consistency
0%

Execution traces reproduce identical state when replayed from checkpoint.

Timeout Recovery
0%

Watchdog-governed processes recovered within 5s SIGTERM escalation window.

Attestation Integrity
0%

SHA-256 environment hashes verified before every execution cycle.

Watchdog Enforcement
0%

Process governance policies enforced with zero orphan process leaks.

Delivery Complete

Execution verified.
Replay preserved.
Delivery complete.

Validated
Tests Pass
Staged
Committed
Replayed
Delivered
Star & Contribute