User repository on GitHub
On-call with TypeSafe Jev: the model judges, plain code decides. Root-cause analysis with no LLM in the loop, and alert triage.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Integrations & Adapters
Jev Oncall
On-call tooling where Jev answers four typed questions per alert - actionable, severity, owning team and duplicate-of - and plain code routes, pages or drops. Includes a root-cause benchmark where Jev and code investigate recorded incidents with no general-purpose LLM in the loop, enforcing budgets and verification.
- Stars
- 4
- Language
- HTML
- License
- MIT
- Published by
- Community
- Last verified
- Oct 2026
About this project
On-call tooling built on Jev, TypeSafe's System One decision model, with one rule throughout: the model judges, plain code decides. Jev answers narrow, typed questions with probabilities; auditable code turns them into actions. Root-cause analysis (current focus): can Jev find why an outage happened, with no general-purpose LLM in the loop? An open benchmark on recorded incidents and on outages adapted from public postmortems, with every trace committed.
Quoted from the project's README. Read the full README.
Highlights
- Pages at P(page) above 0.80 and drops only below 0.20 when actionability agrees; the middle band gets a 15-minute human ack clock.
- Recorded-incident RCA run: Jev found the right service on 57 of 60 cases at about 5.3s per case with zero LLM calls.
- Publication reports a 300-alert triage run at p50 418ms, p95 1477ms and $0.04 per 1,000 alerts.
- Docker Compose demo stages 8 alerts through Prometheus and Alertmanager within a minute.
Watch out
MIT licensed. Pages are not yet delivered to PagerDuty, Slack or Jira; decisions are only exposed via /recent, /pending and the dashboard. RCA results use staged single-cause faults, and the decision rules were revised after seeing results, so several scenarios are no longer held out.
Questions about this entry
Short answers drawn from the linked project or page.
No. Jev only returns probabilities; plain Python turns them into PAGE, REVIEW, TICKET or DROP actions. If Jev errors or times out, the alert falls back to its configured severity, and uncertainty costs a human's attention rather than silence.
Yes. The Docker Compose demo runs without a key and routes every alert by configured severity, which shows the baseline behavior. Adding TYPESAFE_API_KEY enables Jev triage, and the unit tests use a fake Jev with canned probabilities and no network access.
Each production alert gets one Jev call with four typed questions, actionable, severity, owning team and duplicate-of. Code pages above P(page) of 0.80 and drops only below 0.20 when actionability agrees, leaving the middle band a 15-minute human ack clock. The publication reports a 300-alert run at p50 418 ms, p95 1,477 ms and $0.04 per 1,000 alerts.
On 60 recorded faults from RCAEval's Sock Shop and Train Ticket systems, Jev plus code found the right service on 57 of 60 at 5.3 seconds per case with zero LLM calls and about $0.004 per case, against DeepSeek at 53/58 and GLM at 52/54. On the 52 cases all three completed, the counts were Jev 51, DeepSeek 48, GLM 50. A second preregistered run lifted supported diagnoses from 68/120 to 90/120.
Every alternative that ever looked plausible was ruled out by a specific result. That held up: all 9 verified answers in run 1 were right, and 30 across both runs. Unverified is not the same as wrong, since all 17 unverified answers were also right. Its weak spot is evidence: in 22 right answers it cited checks, but none for the injected fault type.
The decision rules were revised after seeing results across v2 to v5, so none of the eight hand-written scenarios is held out any more, and the faults are staged single-cause faults in demo systems with two free-tier LLM baselines. Run it with python3 rca_experiment.py --dry-run --setup alone,jev,jev-agent for the plan, then TYPESAFE_API_KEY and --setup jev-agent --trials 5, and --compare to build an HTML report. MIT licensed.
Not yet. Pages are not delivered to PagerDuty, Slack or Jira, and decisions are exposed only through /recent, /pending and the dashboard. The demo replays a real jev-1.13.0 run over 8 staged alerts, where alert #5 has P(page) of 1.00 but still goes to human review because Jev links it to a checkout incident another team owns.
More Jev resources like this
187.5kGitHub stars
Autonomous agent platform with 187,504 stars for building and running agents from plain English or a visual builder. Its platform features first-class TypeSafe Jev decision blocks for typed routing, filtering, scoring and confidence-gated next-action dispatching.
Jev Integrations & Adapters#community#autonomous-agents#platform
Community
Significant-Gravitas/AutoGPTPython
Significant-Gravitas/AutoGPTPython60.9kGitHub stars
LiteLLM's proxy adds TypeSafe guardrail hooks and a Jev complexity classifier for policy judgments and request routing across 100+ model providers.
Jev Integrations & Adapters#community#python#gateway
Community
BerriAI/litellmPython
BerriAI/litellmPython33kGitHub stars
Alibaba's multi-agent platform, which implements native TypeSafe Jev classification through JevClassifierModel for binary, choice and score routing across agent pipelines. Python 3.11+, Apache-2.0, with a FastAPI agent service and web UI.
Jev Integrations & Adapters#community#multi-agent#framework
Community
agentscope-ai/agentscopePython
agentscope-ai/agentscopePython73GitHub stars
High-performance Go event bus with a bounded MPMC ring buffer, fan-out and ordered handlers, plus an optional routing layer that evaluates deterministic rules first, reuses cached decisions, and asks Jev a typed Choice only for ambiguous cache misses.
Jev Integrations & Adapters#community#go#event-bus
Community
Protocol-Lattice/GoEventBusGo
Protocol-Lattice/GoEventBusGo9GitHub stars
Epistemic runtime for autonomous coding agents that uses Jev as a sub-100ms System 1 gate to shadow-evaluate capsule facts, propose evictions before bloat and prefetch memory between turns. Its fixture run reports 31.3% fewer turns and 52.6% fewer blind discovery operations, with a fail-open circuit breaker.
Jev Integrations & Adapters#community#agents#context-engineering
Community
wojciechwiesner/jit-contextPython
wojciechwiesner/jit-contextPython20.5kGitHub stars
Python agent framework whose TypeSafeModel class runs Jev as a decision model inside Pydantic AI agents: Pydantic output-type fields become typed Jev questions answered together in one request, each answer carrying its own confidence, and tools can be offered as Choice options.
Jev Integrations & Adapters#community#agents#framework
Community
pydantic/pydantic-aiPython
pydantic/pydantic-aiPythonFrom the Jev community
Posts from builders shipping with Jev right now.