TrueFoundry is an enterprise AI platform: model deployment, fine-tuning, serving and an AI gateway in one product. Smartflow is a governance control plane that does not deploy or serve models, and instead governs what happens whenever any application calls any model through any provider. The two overlap at the gateway and diverge everywhere else.
The short version
This is a platform-versus-control-plane comparison rather than a feature bake-off. TrueFoundry's breadth is real and Smartflow does not attempt it. The reason organizations run both is that a control plane governs the AI estate regardless of which platform deployed the model — including the models nobody deployed through a platform at all.
Feature Comparison
| Capability | Smartflow | TrueFoundry |
|---|---|---|
| Primary focus | AI governance control plane. | Enterprise AI platform: deployment, fine-tuning, serving and gateway. |
| Semantic caching | MetaCache resolves a lookup in four phases, first hit wins: intent signature, intent near-miss, exact key, then a VectorLite BERT semantic match. Hit rates are workload-dependent. | Exact-match and semantic caching, documented together. Redis-backed, configured per request by header, with a tunable similarity threshold. |
| Guardrails | Visual policy editor covering PII, topic restriction, jailbreak detection and output moderation, with no code required. | Built-in content moderation, PII and PHI detection, prompt injection, secrets detection, code safety, SQL sanitizer, regex, Cedar and OPA, plus more than twenty third-party integrations. Built-in only on the Developer tier; custom and third-party from Pro. |
| Agent identity | AIDA cryptographic agent credentials, scoped and independently revocable. | Personal and virtual account tokens. A distinct cryptographic agent credential is not a documented feature. |
| Information barriers | Directory group identity with real-time content classification for information barrier enforcement. | Not a documented feature. |
| Regulatory examination | FINRA 3110, FFIEC and EU AI Act evidence packages and model inventory, generated from runtime records. | Audit logging of user actions, resource changes and API activity, with dashboard and API access. SIEM export listed as coming soon. Regulatory evidence generation is not a documented feature. |
| On-premises and air-gapped | Yes. The primary deployment model. | VPC, on-premises and air-gapped deployment, with self-hosting documented in unified and distributed architectures. Enterprise tier only; Developer and Pro are SaaS-only. |
| MCP | MCP JSON-RPC gateway with tool caching. | MCP Gateway supporting remote, virtual, OpenAPI and stdio servers, with centralized credentials, OAuth, audit trails and guardrail enforcement on tool calls. |
| Audit log retention | Customer-controlled. No vendor-imposed ceiling. | Seven days on Developer, thirty on Pro, custom on Enterprise. |
| Model deployment | Not a model deployment platform. Governs rather than deploys. | Full model deployment, fine-tuning and serving. |
The caching row on this page previously read "basic caching, no published benchmarks." That was wrong: TrueFoundry documents exact-match and semantic caching together, with a configurable similarity threshold. Corrected here.
Platform versus control plane
TrueFoundry manages the AI lifecycle. It deploys models, fine-tunes them, serves inference and routes traffic. Smartflow does none of that. It sits between applications and whatever is serving the model, enforces policy on the call, and writes a record of the decision.
The practical consequence is coverage. A platform governs what it deployed. A control plane governs the estate — the models on the platform, the frontier APIs called directly from a notebook, the vendor SaaS with an LLM inside it, and the agent somebody stood up last week. In most enterprises the platform-deployed share is the well-behaved minority, and the governance problem lives in the rest.
That is also why the two compose rather than compete. An organization running TrueFoundry for model operations can put Smartflow in front of TrueFoundry-served models and everything else at the same time, with one policy set and one evidence trail across both.
The tier line matters here
TrueFoundry's pricing makes the deployment question a commercial one before it is a technical one. Developer and Pro are explicitly SaaS-only. VPC, on-premises, air-gapped deployment and data residency are Enterprise. So are custom and third-party guardrails beyond the built-in set, and audit-log retention beyond thirty days.
For a regulated buyer this means the evaluation tier and the deployment tier are different products. A proof of concept on Pro will not exercise the deployment model the organization actually needs, and the guardrail set available in the trial is narrower than the one in production. Worth establishing at the start of a procurement rather than in month three.
Evidence and retention
TrueFoundry's audit logging is competent and honest about what it is: user actions, resource changes, API activity and system actions, emitted as JSON in application logs, filterable by an interceptor tag and exportable to a logging platform of choice. Programmatic access is available to tenant and account administrators. SIEM export is listed as coming soon.
That is an operational audit trail. The gap for a regulated institution is the step after it — turning those records into the specific artifact a supervisor asks for: the inventory of models in use on a given date, the policy in force at the time of a particular call, the supervision record for an agent acting on a client's behalf. Smartflow's examination suite exists to remove that assembly step, and retention on a regulatory timescale rather than a tiered one is a precondition for it.
Where TrueFoundry excels
TrueFoundry offers breadth Smartflow does not attempt: a single platform for deployment, fine-tuning, evaluation, serving and gateway routing, with on-premises and air-gapped options at the enterprise tier and a well-documented self-hosting story in two architectures. The built-in guardrail set is broad and works without external credentials, which matters more in practice than it sounds. Its 2026 release cadence has been strong, including an MCP protocol upgrade, Okta authentication for remote agents, AWS SigV4 support, and a move to OpenTelemetry metrics.
For an organization that wants one platform for both model operations and gateway routing, TrueFoundry is a credible answer and a second product is an argument that has to be made.
When to choose Smartflow
- Governance across the whole estate, not only across what one platform deployed
- Examination evidence: packages and model inventory produced from runtime records
- Regulated-industry controls: agent identity and information barriers, which are not documented TrueFoundry features
- Model infrastructure already settled: where deployment and serving are solved and the missing layer is governance
Common questions
Does TrueFoundry support semantic caching? Yes. Exact-match and semantic caching are documented together, Redis-backed, with a configurable similarity threshold set per request.
Can TrueFoundry run on-premises? Yes, including air-gapped, at the Enterprise tier. Developer and Pro are SaaS-only.
Can both run together? Yes. TrueFoundry for model operations, Smartflow as the governance and evidence layer across TrueFoundry and everything else.
How long does TrueFoundry retain logs? Seven days on Developer, thirty on Pro, custom on Enterprise.
Related reading
AI control plane guide · Information barriers · Examination readiness · Trust Fabric architecture · All comparisons · Model allowlist and denylist · Agentic AI · AI guardrails · Model drift
Verified against each vendor’s published documentation in September 2026. Vendor capabilities change. If something here is out of date, tell us and we will correct it.
TrueFoundry sources: truefoundry.com, truefoundry.com/ai-gateway, truefoundry.com/pricing, truefoundry.com/docs (ai-gateway/caching, ai-gateway/guardrails, ai-gateway/authentication, ai-gateway/gateway-self-hosted, ai-gateway/mcp/mcp-overview, platform/audit-logging, platform/saas-security) and the TrueFoundry changelog.
Put this in the path of your own agents.
Policy enforced inline between your agents and every model and tool they reach, with a record bound to the human who owns it.
Request a Demo Read the docs