agent · model serving
Eidolon AI vs layahost
55 fields both were evaluated on, 24 of them different — including where one discloses something the other does not. Every value links to the document it came from. 31 further fields are disclosed by neither and are listed at the end rather than tabled.
Assurance
| Field | Eidolon AI | layahost |
|---|---|---|
| Runtime governance· | “With kubernetes deployment, you can enforce policies to ensure your agents are deployed with access only to the resources they need.”www.eidolonai.com · checked Aug 1, 2026 | Per-key limits enforced at run time: 600 requests per minute by default, a maximum of 32 questions per request and 32,000 characters of text, with higher limits available on request. These are platform defaults, not customer-configurable policy, and no behaviour on breach is described. “600 requests per minute per key by default, up to 32 questions per request and 32,000 characters of text.”layahost.com · checked Oct 14, 2026 |
| Permission scopes· | “With kubernetes deployment, you can enforce policies to ensure your agents are deployed with access only to the resources they need.”www.eidolonai.com · checked Aug 1, 2026 | no public information layahost.com · checked Oct 14, 2026 |
Agency
| Field | Eidolon AI | layahost |
|---|---|---|
| Human oversight· | no public information www.eidolonai.com · checked Aug 1, 2026 | Borderline answers can be routed down their own branch so that a human reviews what the model is not sure about. “Unsure? Ask a person Borderline answers take their own branch, so a human sees what the model is not sure about.”layahost.com · checked Oct 14, 2026 |
| Goal complexity· | no public information www.eidolonai.com · checked Aug 1, 2026 | Multi-step: decisions are chained on a canvas (the page's example is spam check, intent, routing, urgency), each answer leading to the next question or to an outcome, with one request running the whole flow. “Chain decisions on a canvas: spam check, intent, routing, urgency. Each answer leads to the next question or to an outcome. One request runs the whole flow, and only the questions on the path are billed.”layahost.com · checked Oct 14, 2026 |
| Action space· | no public information www.eidolonai.com · checked Aug 1, 2026 | Constrained classification: returns one of the customer's own defined options with a probability to act on, rather than open-ended text or action-taking. “Anything that ends in a decision: one of your options, with a probability you can act on.”layahost.com · checked Oct 14, 2026 |
| Initiative· | no public information www.eidolonai.com · checked Aug 1, 2026 | Request-driven: work happens per API call. The page describes no self-initiated or scheduled operation. “A call to /v1/decide is one decision; a /v1/systemone request with three questions is three.”layahost.com · checked Oct 14, 2026 |
Safety
| Field | Eidolon AI | layahost |
|---|---|---|
| Data handling· | no public information www.eidolonai.com · checked Aug 1, 2026 | Verdicts and metadata are logged for 7 days by default; raw request text is only logged if the customer opts in. “Logs keep verdicts and metadata for 7 days; request text only if you turn it on.”layahost.com · checked Oct 14, 2026 |
Practicality
| Field | Eidolon AI | layahost |
|---|---|---|
| Pricing model· | no public information www.eidolonai.com · checked Aug 1, 2026 | Usage-based: monthly plans with an included Laya decision volume, or prepaid credit with no subscription. A Laya decision is billed flat regardless of text length; Clef models are billed from credit by input token. “Monthly plans with Laya decisions included, or prepaid credit with no subscription. A Laya decision costs the same whatever the length of the text; Clef models are billed from credit by input tokens.”layahost.com · checked Oct 14, 2026 |
| Price point· | no public information www.eidolonai.com · checked Aug 1, 2026 | Three published per-model prices: Laya at $15 per million decisions; Clef-flash at $0.10 per million tokens (the page reckons about $0.02 per 1,000 short texts); Clef at $0.50 per million tokens (about $0.10 per 1,000 short texts). “latency ~0.1 s price $15 / 1M decisions plans apply Clef-flash clef-flash Cloudflare’s 9B open decision model on Workers AI GPUs. The best value for hard questions. latency 0.4–1 s price $0.10 / 1M tokens ≈ $0.02 per 1,000 short texts Clef clef The 27B version. Usually the most accurate, reads up to 64K tokens of context. latency 0.4–0.5 s price $0.50 / 1M tokens ≈ $0.10 per 1,000 short texts”layahost.com · checked Oct 14, 2026 |
| Availability· | no public information www.eidolonai.com · checked Aug 1, 2026 | Publicly, self-serve available: new accounts start with free credit and no card is required. “Every account starts with $1 of free credit (66.7K decisions), no card needed.”layahost.com · checked Oct 14, 2026 |
| Deployment options· | no public information www.eidolonai.com · checked Aug 1, 2026 | Hosted API only; no self-hosted or on-premise deployment of the service is described. The Laya model runs on the vendor's own servers in Helsinki, Finland (EU). “Laya models run on our servers in Helsinki, Finland (EU).”layahost.com · checked Oct 14, 2026 |
| Integrations· | no public information www.eidolonai.com · checked Aug 1, 2026 | Ships an MCP server for AI agents alongside the HTTP API, and states drop-in compatibility with Google's Perspective API (which the page says shuts down at the end of 2026) and OpenAI's moderation endpoint (beta). “Also a drop-in for Google’s Perspective API , which shuts down at the end of 2026, and for OpenAI’s moderation endpoint (beta). Full API reference and an MCP server for AI agents.”layahost.com · checked Oct 14, 2026 |
| Supported regions· | no public information www.eidolonai.com · checked Aug 1, 2026 | Requests naming the hosted Clef or Clef-flash models are processed by Cloudflare Workers AI, which the vendor states may be outside the EU. The Laya model is processed on the vendor's own EU servers (see deployment_options). “Requests that name a hosted model (clef, clef-flash) are processed by Cloudflare Workers AI, which may be outside the EU and does not use them for training.”layahost.com · checked Oct 14, 2026 |
Foundation models
| Field | Eidolon AI | layahost |
|---|---|---|
| Base models· | GPT4-turbo, MistralLarge, Llama3-8b, ClaudeOpus, ClaudeSonnet “- { apu : GPT4-turbo , title : GPT4 Turbo , default : true } - { apu : MistralLarge , title : Mistral Large } - { apu : Llama3-8b , title : Llama3 8b } - { apu : ClaudeOpus , title : Claude Opus } - { apu : ClaudeSonnet , title : Claude Sonnet }”www.eidolonai.com · checked Aug 1, 2026 | Two model families behind one API: Laya (an open 400M encoder, Apache-2.0, which the vendor hosts) and Cloudflare's Clef and Clef-flash. The page does not say who develops Laya. “Hosted Laya and Cloudflare Clef behind one Jev-compatible API”layahost.com · checked Oct 14, 2026 |
| Model provider· | no public information www.eidolonai.com · checked Aug 1, 2026 | Two providers: the vendor serves Laya from its own EU servers, and Cloudflare supplies the Clef models. “No model wins everywhere. Laya is fast and cheap on our EU servers. Cloudflare’s open Clef models are slower and cost more, and get far more of the hard questions right.”layahost.com · checked Oct 14, 2026 |
| Model swappable· | “You can also define your own agents using vanilla code or any other agent framework you like.”www.eidolonai.com · checked Aug 1, 2026 | Models can be swapped without changing integration code — the same API and response shape is used across Laya and Clef. “Same API, same answers format: change one field, and measure on your own texts.”layahost.com · checked Oct 14, 2026 |
| Open weights· | no public information www.eidolonai.com · checked Aug 1, 2026 | The Laya model is open-weight, released under Apache-2.0. “Open 400M encoder (Apache-2.0) on our CPUs in Helsinki.”layahost.com · checked Oct 14, 2026 |
| Fine-tuning· | no public information www.eidolonai.com · checked Aug 1, 2026 | Customer fine-tuning is offered: the vendor trains a private Laya checkpoint on the customer's labelled examples and serves it on the same API under the customer's own model name, visible only to their account. “We fine-tune a private Laya checkpoint on your labelled examples and serve it on the same API under your own model name, visible only to your account.”layahost.com · checked Oct 14, 2026 |
| Context window· | no public information www.eidolonai.com · checked Aug 1, 2026 | Clef, the 27B model, reads up to 64K tokens of context. No context limit is published for Laya or Clef-flash. “The 27B version. Usually the most accurate, reads up to 64K tokens of context.”layahost.com · checked Oct 14, 2026 |
Ecosystem
| Field | Eidolon AI | layahost |
|---|---|---|
| Protocols supported· | Vendor states API-level compatibility with the Jev API - the same /v1/systemone and /v1/models endpoints, request and response shapes, errors and request IDs, with Jev model names such as jev-latest accepted - and notes the underlying model differs, so probabilities are not identical. “Yes for the API: the same /v1/systemone and /v1/models endpoints, request and response shapes, errors and request IDs, tested with the official Python and JavaScript SDKs. Jev model names such as jev-latest are accepted. The model behind it is different, so probabilities are not identical.”layahost.com · checked Oct 14, 2026 | |
| Tool use· | “Add RAG storage to any application using simple services. Eidolon provides the flexibility to customize RAG to fit your needs.”www.eidolonai.com · checked Aug 1, 2026 | no public information layahost.com · checked Oct 14, 2026 |
| Multi-agent· | no public information layahost.com · checked Oct 14, 2026 | |
| API access· | “Run actions directly through HTTP REST API”www.eidolonai.com · checked Aug 1, 2026 | Self-serve HTTP API: create an account, verify the email address, copy a key and send a request. “Create an account, verify your email for $1 of free credit, copy a key, send one request.”layahost.com · checked Oct 14, 2026 |
| Open source· | Open Source framework “An Enterprise-Grade, Open Source framework”www.eidolonai.com · checked Aug 1, 2026 | The hosted service itself is not offered under an open-source licence. The vendor's "no lock-in" claim refers to the open-weights models it serves (Laya, Clef), which the page says the customer can also run themselves. “The Jev API on open-weights models (Laya, Clef), which you can also run yourself.”layahost.com · checked Oct 14, 2026 |
Disclosed by neither
Both Eidolon AI and layahost publish nothing on these 31 fields. That is a finding about the category rather than a difference between them, so it is recorded here instead of as 31 identical table rows. Each is shown with its source check on the individual profiles.
- Autonomy level
- Operating environment
- User base
- Deployment scale
- Target sectors
- High-risk domains
- Documented incidents
- Support model
- Marketplace presence
- Safety evaluations
- Red teaming
- Safety policy
- Usage restrictions
- Model or system card
- Incident reporting
- Third-party evaluations
- Insurance available
- Insurance carriers
- Coverage limits
- Indemnification
- Liability cap
- Audit trail
- Tamper-evident log
- Explainability
- Compliance certifications
- Regulatory alignment
- Outcome-based pricing
- Settlement mechanism
- Dispute process
- SLA terms
- Evaluation coverage
Questions
- How do Eidolon AI and layahost compare on human oversight?
- Eidolon AI: no public information disclosed. layahost: Borderline answers can be routed down their own branch so that a human reviews what the model is not sure about..
- How do Eidolon AI and layahost compare on goal complexity?
- Eidolon AI: no public information disclosed. layahost: Multi-step: decisions are chained on a canvas (the page's example is spam check, intent, routing, urgency), each answer leading to the next question or to an outcome, with one request running the whole flow..
- How do Eidolon AI and layahost compare on action space?
- Eidolon AI: no public information disclosed. layahost: Constrained classification: returns one of the customer's own defined options with a probability to act on, rather than open-ended text or action-taking..
- How do Eidolon AI and layahost compare on initiative?
- Eidolon AI: no public information disclosed. layahost: Request-driven: work happens per API call. The page describes no self-initiated or scheduled operation..
- How do Eidolon AI and layahost compare on pricing model?
- Eidolon AI: no public information disclosed. layahost: Usage-based: monthly plans with an included Laya decision volume, or prepaid credit with no subscription. A Laya decision is billed flat regardless of text length; Clef models are billed from credit by input token..
- How do Eidolon AI and layahost compare on price point?
- Eidolon AI: no public information disclosed. layahost: Three published per-model prices: Laya at $15 per million decisions; Clef-flash at $0.10 per million tokens (the page reckons about $0.02 per 1,000 short texts); Clef at $0.50 per million tokens (about $0.10 per 1,000 short texts)..