Agentic Index

AI agent platforms that let you bring your own model or route between providers (2026)

135 of the 566 agentic AI platforms in the Agentic Index document bringing your own model, bringing your own key or routing tasks between providers in full. That is 23.9% of the platform field. Another 125 document some model choice without enough evidence to clear, 304 document none at all, and 2 could not be verified either way, which accounts for all 566. This is a bar, not a leaderboard: a platform either documents model choice as a buyer facing surface in its own public materials or it does not clear, and Partial evidence does not count. Every platform that clears is named below, ordered by total coverage across all 14 capabilities.

The finding underneath the list is the part a buyer cannot get anywhere else. Model independence and enterprise governance travel together, and are still rare together. Only 37 of the 566 platforms in the index, 6.5%, document both model choice and the four capability enterprise governance bar used on the enterprise security and compliance page. A team that needs both is shopping that field, not this one.

It would be easy to read that as a trade off, and it is not one. The association runs the other way. Documenting model choice makes a platform about twice as likely to clear enterprise governance, not less likely. Across the whole pool, 77 platforms clear governance, a base rate of 13.6%. Among the 135 platforms on this page the rate is 27.4%. It holds in reverse too: 48.1% of the 77 governance cleared platforms document model choice, against a field base rate of 23.9%. If the two capabilities were independent the overlap would be about 18 platforms. It is 37. The vendors that expose the model are, on documented evidence, the same vendors that expose the audit log and the deployment target.

So the scarcity sits in the joint requirement rather than in any tension between the two, and it is real scarcity. 98 of the 135 platforms on this page still do not clear governance, and 40 of the 77 governance cleared platforms still do not document model choice. Doubling a small number leaves a small number. A buyer who has already shortlisted on governance has ruled out most of this list without meaning to, and a buyer who shortlisted on model independence has done the same in reverse, which is why the two requirements are worth stating together at the start of an evaluation rather than discovering in sequence.

The same shape shows up on deployment, which is the narrower question most buyers actually mean. 82 of these 135 platforms also run inside your own boundary, well above the pool base rate, and 53 will take your model while the software keeps running in the vendor's cloud. Which of those two shapes is the right answer depends on whether your constraint is which model sees the data or where the software runs, and the two are separate purchases. The self hosting page counts the same intersection from the deployment side, where the gap runs the other way.

The bar, and what the 135 platforms that clear it also document

Model flexibility and routing at Full is the whole bar on this page: the platform documents bringing your own model or bringing your own key (BYOK), or routing tasks between models or providers, rather than being locked to one vendor hosted model. The table below is not the bar. It shows how the 135 platforms that clear it score on enterprise governance, next to the rate across the whole pool.

Governance capability	What has to be documented	Pool	Also on this list
Security and identity governance	SOC 2 or ISO 27001, single sign on (SSO or SAML) and role based access control (RBAC)	277 (49%)	86 (64%)
Deployment and data residency	self hosted, on premises, air gapped or private VPC deployment, or documented control over where agent data sits	188 (33%)	82 (61%)
Observability and auditability	tracing and audit logging across agent runs	319 (56%)	87 (64%)
Human oversight and guardrails	a human approval step before an agent action executes	281 (50%)	75 (56%)

Full means the vendor publishes evidence meeting the capability in its own public materials, under the Agentic Index verification standard. Pool counts the 566 platforms in the index. Also on this list counts the 135 that clear the model bar and document that capability too. Every row runs higher in the right hand column than in the left, which is the positive association stated above holding capability by capability rather than only on the combined bar. One honest limit on the bar itself: this axis grades bring your own key and routing between providers as a single capability, so a Full grade does not by itself separate the two. Read the entry and the vendor profile behind it before assuming which one a platform documents.

Clears the bar and scores 12.5 or higher of 14 overall

These 13 platforms document model flexibility and routing in full and also sit at the top of the Agentic Index coverage score across all 14 capabilities. Ordered by total coverage, ties broken alphabetically.

1. 1.UiPath

   13.5 / 14 capabilities

   enterprise operations agents, multi-agent platforms, agent builders

   Agentic automation orchestrating agents, robots and people end to end. Autopilot supports GPT, Gemini, Claude and any compatible model, run on the customer's cloud of choice or their own infrastructure, with an LLM selector in Agent Builder. The highest verified coverage score in the index and it does not tie you to a model.

2. 2.Automation Anywhere

   13.0 / 14 capabilities

   enterprise operations agents, multi-agent platforms, agent builders

   Agentic process automation pairing goal driven agents with RPA bots. Model agnostic by design rather than by configuration: it connects to standard, fine tuned and RAG foundation models from Amazon Bedrock, Google Vertex AI, Azure OpenAI and OpenAI rather than hosting its own.

3. 3.CrewAI

   13.0 / 14 capabilities

   multi-agent platforms

   Thinner evidence than the score suggests. Both the model row and the wider grade rest on the internal capability research report rather than a named CrewAI documentation page, so the basis is thinner than the enterprise entries beside it even though the score is the same.

   Open source multi agent orchestration for collaborative crews, MIT licensed and Python native, so the model choice is yours in the same sense the code is. Read the evidence caveat beside this entry before treating the grade as thick.

4. 4.Salesforce

   13.0 / 14 capabilities

   enterprise operations agents

   CRM platform whose Agentforce layer runs agents across sales, service and marketing. The Atlas Reasoning Engine supports OpenAI, Anthropic and Google Gemini, so an enterprise can pick the model for a governed deployment rather than inherit one. It is also one of the platforms that clears the full governance bar.

5. 5.Agno

   12.5 / 14 capabilities

   agent infrastructure platforms

   Thinner evidence than the score suggests. The model row cites the internal expansion evidence report rather than a named Agno documentation page. The containment story is strong on licence terms, the model evidence is thinner than the score suggests.

   High performance agent runtime, formerly Phidata, Apache 2.0 licensed, which is the strongest structural position for model independence because the runtime is yours to point anywhere. The evidence caveat beside this entry applies to the model row specifically.

6. 6.Atomicwork

   12.5 / 14 capabilities

   enterprise operations agents

   AI native ITSM and ESM platform where every AI coworker gets a role, a budget, scoped access and an audit trail. It lets teams bring their own agent harness and choose from OpenAI, Anthropic and Gemini on an ensemble multi model architecture, which is unusual candour for a product sold as an outcome.

7. 7.Boomi

   12.5 / 14 capabilities

   enterprise operations agents, multi-agent platforms, agent builders

   Integration platform with an agent control tower over enterprise systems. Model agnostic with MCP support across Amazon Bedrock, Snowflake Cortex and OpenAI, with roadmap support for custom large and small models, and the integration layer is where its evidence comes from.

8. 8.Coworker

   12.5 / 14 capabilities

   enterprise operations agents, agent builders

   Enterprise agent platform built on an organizational memory knowledge graph. The most complete routing story in the headline tier: intelligent routing scores each task across closed, open source and self hosted models including Llama and Mistral, adopts new models automatically, and still allows per task manual control.

9. 9.CrowdStrike

   12.5 / 14 capabilities

   multi-agent platforms, agent builders (secondary lane membership, primary category Security / SOC agent)

   Security platform whose Charlotte AI runs agentic detection, triage and response, with AgentWorks offering model optionality across Anthropic, OpenAI and NVIDIA plus Bedrock and SageMaker infrastructure. It qualifies here on secondary lane membership rather than as a primary platform.

10. 10.Pydantic AI

    12.5 / 14 capabilities

    agent infrastructure platforms

    Open source MIT licensed Python agent framework, and on documented evidence the widest model surface on this page: compatible with virtually every provider with a documented path for custom models, plus a gateway adding cost control, routing and bring your own key alongside built in providers.

11. 11.Rasa

    12.5 / 14 capabilities

    multi-agent platforms

    Open source conversational AI with an enterprise framework, self hosted or SaaS, with multi model support documented in the open source docs. Read it as a posture backed by code you can run rather than a configurable enterprise routing product.

12. 12.Sim

    12.5 / 14 capabilities

    multi-agent platforms, agent builders

    Open source Apache 2.0 agent workspace and the sharpest single instance on this page: fifteen plus providers documented, bring your own key at base provider pricing with no markup, local models through Ollama or VLLM at no API cost, and enterprise permission groups that restrict which providers are allowed.

13. 13.SimplAI

    12.5 / 14 capabilities

    multi-agent platforms, agent builders, agent infrastructure platforms

    Enterprise agentic AI operating system spanning cloud, customer cloud, VPC, on premises and air gapped with no architectural change between them. A model layer orchestrates across LLMs with routing and evaluation and the model is selected per agent. One caveat a buyer should carry: bring your own models and fine tuning are documented but gated to the enterprise tier.

The remaining 122 platforms that clear the bar

Every one of these documents model flexibility and routing in full. They score below 12.5 of 14 on total coverage, which says something about breadth across the whole taxonomy rather than about how open they are on models. Several of the strongest model positions in the index are narrow runtimes and gateways that score low precisely because they do one thing.

Common questions

Which AI agent platforms let you bring your own model or route between models?

135 of the 566 agentic AI platforms in the Agentic Index document model flexibility and routing in full, meaning bring your own model, bring your own key, or routing tasks between models and providers. That is 23.9% of the platform field. The strongest documented positions by total capability coverage are UiPath, Automation Anywhere, CrewAI and Salesforce, followed by Agno, Atomicwork, Boomi, Coworker, CrowdStrike, Pydantic AI, Rasa, Sim and SimplAI. Another 125 platforms document some model choice without enough evidence to clear, 304 document none, and 2 could not be verified either way. Graded from public evidence only.

Can you get model choice and enterprise governance in the same platform?

Yes, and the two go together more often than chance, but the combination is still uncommon. 37 of the 566 platforms in the pool document both model flexibility and the four capability enterprise governance bar of security and identity, deployment and residency, observability, and human approval before an agent acts. That is 6.5% of the field. The association is positive in both directions: platforms documenting model choice clear governance at 27.4% against a field base rate of 13.6%, and platforms clearing governance document model choice at 48.1% against a base rate of 23.9%. Each roughly doubles the odds of the other, and if they were independent the overlap would be about 18 platforms rather than 37. The scarcity is in the joint requirement, not in a conflict between the two.

What counts as bring your own model on this page?

Documented public evidence that model choice is a buyer facing surface: a provider list you can select from, bring your own model or bring your own key with your own provider account, or a routing policy that sends tasks between models. Partial evidence, meaning one alternative provider or a roadmap statement, does not clear. One honest limit worth knowing: the axis grades bring your own key and routing between providers as a single capability, so a Full grade does not by itself tell you which of the two a given platform documents. Read the entry and the linked vendor profile for that.

If a platform takes your own model, does it also run in your own environment?

More often than not. 82 of the 135 platforms here also document self hosted, on premises, air gapped or private VPC deployment in full, which is 60.7% against a pool base rate of 33.2%. The other 53 will take your model while the software keeps running in the vendor's cloud. That is a real answer for a team whose constraint is which model sees the data rather than where the software runs, and the wrong answer for a team whose constraint is the boundary.

Does a platform missing from this list mean it locks you to one model?

It means the vendor does not document model choice as a buyer facing option in its own public materials. Plenty of platforms swap models internally. The point is that a buyer could not verify it before signing, and could not choose after. Unable to verify is a statement about the evidence, not an accusation about the product.

Is this ranking paid or sponsored?

No. No vendor pays for placement, no vendor has reviewed this page, and every grade comes from the vendor's own public materials under the Agentic Index verification standard. 988 vendors are graded against the same 14 capabilities. Data last verified August 22, 2026.

Method: membership is the same 566 platform pool used by the best agentic AI platforms in 2026, drawn from 988 researched vendors. The editorial pages over this pool are one method with different bars, not several opinions. Every grade comes from the vendor's own public materials under the Agentic Index verification standard. No vendor pays for placement and no vendor has reviewed this page. Data last verified August 22, 2026. How this evidence is graded

Related: platforms you can self host or run in a private VPC, enterprise security and compliance platforms, platforms that support Model Context Protocol tools, how all 988 vendors score on model flexibility, with the rate in each of the 13 use case lanes, compare platforms side by side.