Free Visit Website
Tracked since 2026
The Bottom Line
- Price
- Free, no paid tier. Compare plans
Key facts
- An 11 MB open-weight decision model that runs on CPU and works offline, with no API or GPU required.
- Returns probability distributions for every option, enabling threshold-based automation and easy escalation.
- Supports choice, yes/no (noul), and score questions, and can adapt to new label mixes without retraining.
Pros
- Extremely small footprint (11 MB) and runs on CPU, making it easy to ship inside any service
- Open source with open weights, no per-call costs, and full offline capability
- Provides calibrated probabilities for threshold-based automation, reducing manual review
Cons
- Only supports English; not suitable for multilingual workflows
- Accuracy on brand-new tasks without labeled data is low; requires a few dozen examples to validate
- Not designed for mobile devices (requires Python and PyTorch on CPU)
What is ej?
ej is a compact decision model library (11.4 MB) designed for routing, triage, and scoring tasks. It runs entirely on CPU, works offline, and is distributed as a single file with open weights under Apache-2.0. Developers use it to automate decisions like ticket routing, urgency scoring, and yes/no checks by passing a state (text or JSON) and custom questions. The model returns a probability distribution for each option, enabling threshold-based automation without parsing strings or making API calls. It includes an adapt method to adjust to a specific workflow's label mix from a handful of labeled records without retraining the weights. ej is best suited for repeated, well-defined decision tasks where you can provide a few dozen labeled examples to verify accuracy.
Louis Corneloup Source hitl.systems ↗
Key Features
- Single 11.4 MB file containing encoder, heads, and tokenizer
- Runs on CPU, works offline, no external dependencies at load time
- Returns a probability for every option in one pass
- Supports three question types: choice, noul (yes/no), and score (ordered levels)
- Adapt method learns workflow label mix from a few labeled records without changing weights
- Open weights and Apache-2.0 licensed code with SHA-256 integrity check
- No JSON parsing or retries needed; probabilities are directly actionable
Pricing
Free
ej is completely free to use with no hidden costs.
Reviews
$99 Free with your review
Write a reviewReview ej, get a free AI guide
Share your experience and we will send you Improve Your Thinking Patterns Using ChatGPT, free.
Best ej Alternatives
Top alternatives based on features, pricing, and user needs.
Redlight Greenlight for Claude Code Free
Respond to Claude Code Permissions From Anywhere
Node-RED Free
Flow-based programming for IoT
Composio MCP Freemium
Empower AI agents to act across 1000+ apps securely
CrewAI Freemium
Multi-agent orchestration framework for autonomous AI teams
Trigger.dev Freemium
Background jobs for serverless
Val Town Freemium
Social website for writing code
Still deciding?
Most buyers shortlist 2 or 3 tools before committing. Pull a side-by-side comparison or browse the full alternatives shortlist below.
Explore More
ej FAQ
How does ej automate ticket routing?
ej automates ticket routing by accepting a state (text or JSON) and custom questions, then returning a probability distribution for each routing option. This allows developers to set thresholds for automation without parsing strings or making API calls, using a compact offline model.
How does ej compare to Node-RED for decision automation?
Unlike Node-RED, which is a flow-based programming tool for wiring together hardware and APIs, ej is a compact decision model library that runs entirely on CPU and offline. ej provides calibrated probabilities for threshold-based automation, whereas Node-RED requires building custom logic for each decision. ej's single-file distribution and open weights make it simpler to embed in services.
What are the limitations of ej for multilingual workflows?
ej only supports English and is not suitable for multilingual workflows. Its accuracy on brand-new tasks without labeled data is low, requiring a few dozen labeled examples to validate. Additionally, ej is not designed for mobile devices as it requires Python and PyTorch on CPU.
What kind of user benefits most from ej's compact decision model?
Developers working on repeated, well-defined decision tasks like ticket routing, urgency scoring, or yes/no checks benefit most from ej. Teams that need offline, CPU-only automation without per-call costs and can provide a few dozen labeled examples to verify accuracy will find ej a good fit.
How is ej priced?
ej is free to use with no paid plan required. It is distributed as a single file with open weights under Apache-2.0, so there are no per-call costs and it runs fully offline.
Can ej adjust to a specific workflow's label mix without retraining?
Yes, ej includes an adapt method that adjusts to a specific workflow's label mix from a handful of labeled records without retraining the weights. This allows developers to fine-tune the model for their particular use case using only a few dozen examples.
Does ej require an internet connection or API calls to automate decisions?
No, ej runs entirely offline on CPU and does not require any internet connection or API calls. It is distributed as a single file, so developers can embed it directly into their services for decision automation.
How does ej return decision results for threshold-based automation?
ej returns a probability distribution for each option, enabling threshold-based automation without parsing strings. Developers can set confidence thresholds to trigger actions, reducing the need for manual review.
Source: hitl.systems