# Alternatives to IrisAgent > Source: https://theagentsindex.com/irisagent/alternatives (current datasheets, evidence-ranked) Candidates share the baseline category or are previously researched peers. Requirements, preserved strengths, explicit trade-offs and research coverage decide the order; paid tier does not. ## Shortlist ### 1. Ada Enterprise AI customer service agents across voice, chat, email and messaging, with tools that act in your systems. **Research coverage:** 35/35 criteria **Where it improves on IrisAgent:** - [Agent permissions: yes — Every discovered tool starts disabled until an admin enables it, write tools are recommended behind a controlled playbook step, and an ASK step confirms with the customer before the RUN step executes.](https://docs.ada.cx/docs/automation/tools/mcp-tools) - [Open source: partial — Only the client-side chat UI library is distributed publicly on npm as @ada-cx/lovelace; the platform and Reasoning Engine are hosted with no published source.](https://docs.ada.cx/lovelace/overview) - [Seat model: partial — Dashboard access is per named user with four roles, invited and removed per Ada instance, but no seat price, seat minimum or seat ceiling is published anywhere, so the commercial half of the seat model is unknown.](https://docs.ada.cx/docs/other/team-access/my-team) **Trade-offs against IrisAgent:** - [Model choice: no — Ada picks the models: it curates the portfolio, benchmarks candidates on accuracy, safety, latency and cost, and phases in updated versions itself, with no documented customer model selection.](https://www.ada.cx/platform/trust/) - [Self-hosted: no — Every documented deployment is a vendor-hosted Ada instance reached at its own subdomain, with Ada holding the conversation data under its own retention schedules; no self-hosted or bring-your-own-infrastructure option appears anywhere in the documentation.](https://docs.ada.cx/docs/other/data-retention) - [Free tier: no — There is no self-serve sign-up or free plan: the pricing route is a demo booking form, and the vendor's own process is an expert review, a custom demo and a business case.](https://www.ada.cx/pricing/) [View Ada](https://theagentsindex.com/ada) ### 2. Maven AGI Enterprise AI agent platform that resolves customer conversations across chat, email, voice and internal tools. **Research coverage:** 33/35 criteria **Where it improves on IrisAgent:** - [Agent permissions: yes — Permissions are configured centrally: one control layer sets guardrails, escalation patterns and which systems an agent may act in, and Charters scope behaviour and access per situation.](https://www.mavenagi.com/product/agent-designer) - [Admin controls: yes — Administrators constrain the agent directly: routing rules, workflow policies, identity controls and deterministic logic gate every action, and Agent Designer controls which systems the agent may act in.](https://www.mavenagi.com/features/trust-compliance) - [Audit log: yes — Agent decisions and actions are logged, traceable and exportable with SIEM integration, and the vendor separately lists comprehensive audit logging among its platform security controls.](https://www.mavenagi.com/features/trust-compliance) **Trade-offs against IrisAgent:** - [Open models: no — The agreement enumerates its LLM providers as OpenAI and Anthropic and promises written notice if any other is added, a closed list with no open-weight or self-hosted model in it.](https://www.mavenagi.com/legal/msa) - [Self-hosted: no — The vendor's own marketplace listing classes delivery as SaaS and states the software is not deployed in the buyer's AWS account; no on-premise, VPC or single-tenant option is published on any tier.](https://aws.amazon.com/marketplace/pp/prodview-l2djyrnxnfffq) - [How it meters: partial — The vendor names resolution volume as what the price scales with, but publishes no per-resolution rate; the contract meters against a subscription usage allowance with overage priced privately in the order form.](https://aws.amazon.com/marketplace/pp/prodview-l2djyrnxnfffq) [View Maven AGI](https://theagentsindex.com/maven-agi) ### 3. Zendesk AI agents Customer-service AI agents for messaging, email and voice, billed per automated resolution. **Research coverage:** 35/35 criteria **Where it improves on IrisAgent:** - [Agent permissions: yes — Admins set policies and guardrails, review the decision path, and every interaction is scored by built-in QA; Enterprise adds approval workflows.](https://www.zendesk.com/service/ai/ai-agents/) - [Seat model: yes — Priced per agent seat per month with no published seat floor; the resolution allowance is also derived per seat (5 ARs/agent/month on Team and Suite Team, 10 on Professional).](https://www.zendesk.com/pricing/) - [Admin controls: yes — Reasoning controls on every plan, automated QA scoring of each interaction, and Enterprise approval workflows.](https://www.zendesk.com/service/ai/ai-agents/) **Trade-offs against IrisAgent:** - [Claude: partial — Claude is not what powers the agent; it is reachable as an external action in action flows using your own Anthropic key.](https://support.zendesk.com/hc/en-us/articles/10630133525018-Using-Claude-actions-in-action-flows) - [Open models: no — Every generative path the vendor enumerates is a managed hosted API; no open-weight or customer-supplied model is offered.](https://www.zendesk.com/trust-center/) - [Model choice: no — Zendesk selects the models: generative LLMs plus its own proprietary intent models. No model picker is documented.](https://www.zendesk.com/service/ai/ai-agents/) [View Zendesk AI agents](https://theagentsindex.com/zendesk-ai-agents) ### 4. Decagon Enterprise platform for building and running AI customer-support agents across chat, voice and email. **Research coverage:** 31/35 criteria **Where it improves on IrisAgent:** - [Agent permissions: yes — The agent executes real operations such as refunds and identity verification, and how far it may go before escalating is set by AOP guardrails and guidelines.](https://decagon.ai/product/overview) - [Admin controls: yes — Admins constrain the agent two ways: role-based platform permissions, and AOP guardrails that gate refunds, escalations and brand voice.](https://decagon.ai/security) - [Single sign-on: yes — SSO is offered against enterprise identity providers, with Okta and Microsoft Entra named; the underlying SAML or OIDC protocol is not stated.](https://decagon.ai/security) **Trade-offs against IrisAgent:** - [Claude: partial — Anthropic is named as one of Decagon's AI providers and is listed as an LLM subprocessor, but the buyer never selects Claude — routing is Decagon's.](https://decagon.ai/security) - [GPT: partial — OpenAI is named as an AI provider and appears as an LLM subprocessor on the trust centre, yet no page offers the customer a choice of GPT.](https://decagon.ai/security) - [Model choice: no — Model selection belongs to Decagon, not the buyer: the engine is proprietary and most traffic runs on models Decagon trained itself, with no published model picker.](https://decagon.ai/blog/introducing-decagon-labs) [View Decagon](https://theagentsindex.com/decagon) ### 5. Sierra Enterprise platform for customer-service AI agents across voice, chat, email and messaging, priced by outcome. **Research coverage:** 32/35 criteria **Where it improves on IrisAgent:** - [Agent permissions: yes — Autonomy is configurable per workflow: you set goals and guardrails and name the steps that require human sign-off before the agent proceeds.](https://sierra.ai/product/horizon) - [Admin controls: yes — Administrators constrain the agent through goals, guardrails, topic filters and human sign-off points defined before anything reaches customers.](https://sierra.ai/product/trust-and-reliability) **Trade-offs against IrisAgent:** - [Claude: partial — Anthropic is a named AI-services subprocessor, so Claude is in the mix, but Sierra routes models itself and never exposes a Claude selector.](https://trust.sierra.ai/subprocessors) - [GPT: partial — OpenAI is listed as an AI-services subprocessor, so GPT models are used under the hood, chosen by Sierra's router rather than by the buyer.](https://trust.sierra.ai/subprocessors) - [Model choice: no — Model selection belongs to Sierra: it routes each turn to what it judges optimal and swaps providers on outages, with no published customer model choice.](https://sierra.ai/product/trust-and-reliability) [View Sierra](https://theagentsindex.com/sierra) ### 6. Fin Intercom's customer agent that resolves support conversations across chat, email, voice and social channels. **Research coverage:** 35/35 criteria **Where it improves on IrisAgent:** - [Agent permissions: yes — The buyer sets the boundary: Procedures apply business rules and can require approvals, data access can be restricted to named fields, and Fin escalates rather than answering when confidence thresholds fail.](https://www.intercom.com/security) - [Seat model: partial — Fin itself carries no seat cost and the vendor states teammates are unlimited, but the floor is a 50-outcome monthly minimum, and pairing Fin with the Intercom helpdesk adds a per-seat charge for Full seats.](https://fin.ai/pricing) - [Admin controls: yes — An administrator constrains the agent directly — restricting which fields Fin may read, renaming internal values, controlling integration permissions through OAuth, and masking card numbers automatically.](https://fin.ai/integrations) **Trade-offs against IrisAgent:** - [Claude: partial — Anthropic is one of the providers Fin falls back to for resilience, but the answer model is Intercom's own Apex and the buyer is not offered a model picker, so Claude is under the hood rather than selectable.](https://fin.ai/trust-reliability) - [GPT: partial — OpenAI models are part of the same automatic failover pool as Anthropic and Google, so GPT is used by the platform but is not a model the customer chooses or configures on any tier.](https://fin.ai/trust-reliability) - [Open models: no — The enumerated provider list is OpenAI, Anthropic, Google and Intercom's own models; no open-weight option is offered.](https://fin.ai/trust-reliability) [View Fin](https://theagentsindex.com/intercom-fin) ## Method Unknown data is never treated as a benefit or a trade-off. Featured placement is labelled but cannot alter recommendation order.