The inspiration for our AI startup stemmed from the founder's personal experience and passion for efficient problem-solving. He recognized a gap in the market for effective tools to analyze, plan, execute, and monitor personal tasks.To address this, he set out to create an accessible solution that integrated various scientific disciplines and open-source technologies.The goal is to provide individuals with invaluable clarity and improved effectiveness at problem-solving in their daily lives.
A collection of specialized machine-learning tools as modular components. By hooking the components together, the pipeline approach can solve for a specific task none of the programs was designed for.By orchestrating a swarm of AI agents as a single entity, a human operator can achieve enhanced performance on increasingly complex tasks that would be physically unfeasible with conventional methods.
Backed by empirical evidence from cognitive science, neuroscience, psychology and artificial intelligence.By leveraging cognitive architectures, mental models and AI frameworks, you can model complex systems, recognize patterns, and make informed decisions.With evidence-driven advanced reasoning, you can significantly reduce cognitive biases, improve decision-making accuracy, and deliver expected outcomes.
The vast, untapped reservoir of publicly available information offers a wealth of data and insights from diverse sources, including open-source data repositories and collaborative platforms.By harnessing this intelligence, you can leverage the collective knowledge of experts and communities, identify patterns and connections that may have gone unnoticed, and inform decision-making with data-driven insights.The potential for breakthrough is vast.
Orchestrating multiple intelligent agents can revolutionise problem-solving by unlocking unprecedented levels of efficiency, innovation, and accuracy.By combining the strengths of various agents, you can rapidly explore solution spaces, identify optimal outcomes, driving faster solution discovery and unlocking the full potential of collective intelligence.
A suite of evaluation on our Phoebe series of models.
* As of 2024-09-27
| Model | Average Score | Type |
|---|---|---|
| gpt-4-turbo-2024-04-09 | 86.35 | Closed Source |
| claude-3-opus-20240229 | 82.19 | Closed Source |
| Meta-Llama-3-70B-Instruct | 82.13 | Open Source |
| Qwen2.5-14B-Instruct | 79.23 | Open Source |
| Phi-3-medium-4k-instruct | 76.34 | Open Source |
| Phi-3-small-8k-instruct | 73.49 | Open Source |
| Meta-Llama-3-8B-Instruct | 68.88 | Open Source |
| mistral-v02-instruct | 68.18 | Open Source |
| Meta-Llama-3.2-11B-Vision-Instruct | 67.88 | Open Source |
| inervi-phoebe | 77.04 | Open Source |
Our models are within 89% performance of closed source models, utilising only a fraction of their parameters.
Passkey Retrieval Task

RULER Evaluation on Long Context Tasks
* Source: "https://github.com/hsiehjackson/RULER"
| Model | 4k | 8k | 16k | 32k | 64k | 128k |
|---|---|---|---|---|---|---|
| llama-2-7b(baseline) | 85.6 | - | - | - | - | - |
| gpt-4-1106-preview | 96.6 | 96.3 | 95.2 | 93.2 | 87.0 | 81.2 |
| meta-llama-3-70b-instruct | 95.4 | 94.7 | 93.2 | 85.9 | 22.5 | - |
| phi-3-mini-128k-instruct | 86.7 | 78.1 | 75.6 | 70.3 | 58.9 | 43.3 |
| mistral-v02-instruct | 93.6 | 91.2 | 87.2 | 71.5 | - | - |
| inervi-phoebe | 87.8 | 87.2 | 87.7 | 69.0 | 46.8 | 19.0 |
| Metric | Description |
|---|---|
| Faithfulness(F) | Measures how factually accurate is the generated response. |
| Type | Description |
|---|---|
| no-retrieval | Baseline generation without any external knowledge. |
| sample-retrieval | Augmented generation with sample passage(1x) around 300 words in length. |
| hybrid-retrieval-rerank | Augmented generation with candidate passages(5~10) from dense and sparse retrievers, followed by reranking step. |
| hybrid-retrieval-rerank-multi-step | Augmented generation with hybrid retrieval from queries with varying complexities. |
Simple fixed-rate monthly plans(SGD) to meet budget needs.
1. On-device models only
2. 2,000 effective context length
3. 1,000 input
4. 1,000 output
5. Security provided by Cloudflare
6. Support : community*
7. Ideal for experimentation
*Community support is based on best-effort basis
1. All from "Basic"
2. Multi-agents(Built with Llama and friends)
3. 16,000 effective context length
4. 2,000 input
5. 1,000 output
6. Hybrid retrieval augmented generation(RAG) with vectors and knowledge graph
7. Memory(chat history)
8. Tools usage(internet access, math, document parsing and more)
9. Support: service desk(24~48h response)
10. Ideal for latency-insensitive tasks
1. All from "Standard"
2. Unlimited access*
3. Ideal for latency-sensitive tasks
*Unlimited access is rate-limited after thresholds
FAQ
What is Inervi?Inervi is a platform designed to accelerate individual creative problem-solving.Which desktop browsers are supported?Support is only extended to 64-bit latest stable version of Chrome(Linux, Windows build). Other browsers and their derivatives might work (Chromium, Edge, Firefox and Safari).Will Inervi be available on mobile platforms like iOS and Android?No, Inervi is a proof-of-concept. Resources are targeted to validating as a minimum viable product. However, the web client can be accessed using mobile Chrome from v108 onwards.Is the Basic plan forever free?Yes, we do not up-sell or automatically convert your plan without getting your explicit consent.How is the billing cycle calculated?A subscription's billing date is determined by two factors: billing cycle anchor which defaults to when the subscription was created and recurring interval which defaults to monthly.How do I upgrade or downgrade subscriptions?You can access the customer portal via account page to manage an active subscription.Can I use multiple accounts?Yes, you can subscribe to multiple plans under the same email address and each plan will have its own independent billing cycle.Is my subscription prorated when changed in the middle of the billing cycle?Yes, if you upgrade from 5 SGD per month to 10 SGD per month option, you will be charged prorated amounts for the time spent on each option.Assuming the change occurred halfway through the billing period, you will be billed an additional 5 SGD: -5 SGD for unused time on initial price, and 10 SGD for the remaining time on the new plan.What happens to my remaining balance if I downgrade/cancel the subscription before the next billing cycle?Every customer has a credit balance where adjustments could be a credit or a debit operation. These adjustments sum up to a balance that Inervi can apply to future invoices.What happens when I delete my account before my subscription expires?Your access to the paid subscription will last until the end of its period. Access is restricted to the same email you used during the subscription setup.Why are verification emails not reaching my custom domain?Your DNS settings may be misconfigured or the appropriate MX record doesn't exist or has to replicate to all available Domain Name System(DNS) servers around the world. This could take up to
72 hours.Which region(s) are supported?Singapore(SG).
For all related enquiries, feedback and feature requests, please email us.