TrueFoundry
Start by wiring your first agent the way you’d set up a microservice: choose where it runs, which models it can call, and who can use it. Point the platform at your cloud, VPC, or data center, then add your model providers and in-house checkpoints. Set route rules to prefer certain models by task, attach secrets securely, and define usage limits so teams don’t blow through budgets. Grant access with roles tied to your identity provider, and decide where logs and traces land. With this done, you have a controlled entry point for all agent traffic and a single place to manage retries, timeouts, and failover.
Next, assemble an agent for a real task—say, customer support triage. Register the tools it needs: a ticketing API, your knowledge base, a billing service, and a redaction utility. The tool registry enforces input/output contracts so malformed calls are rejected before anything breaks downstream. Add stateful recall so the agent remembers prior steps and decisions, and enable step-by-step planning to decompose complex requests into smaller actions. Connect retrieval to your docs with configurable chunking and ranking. Run the agent in a sandbox, watch the action graph, and iterate on the tool signatures until success rates look solid. When ready, promote to production with traffic splitting to keep a safety net.
Keep control of language behavior with prompt lifecycle workflows. Store prompts with immutable versions, name them by intent (e.g., “refund-triage:v3”), and track changes alongside evaluation scores. Run A/B tests across models and prompts using the same dataset of conversations or user tasks. Monitor hit rates, latency, cost per request, and tool success by step. If quality drops, roll back a version or pin a known-good combo for critical paths. Set guardrails that auto-sanitize inputs, block unsafe tool calls, and redact sensitive fields before any third-party hop. Alerts and dashboards make it easy to catch regressions early and prove performance over time.
Operate at scale with enterprise controls. Enforce data residency per project, cap spend by team, and set rate limits by API key. Every action is captured in an immutable audit trail, and granular permissions keep model access, prompts, tools, and datasets scoped to the right people. Deploy agents close to your data for lower latency while meeting SOC 2, HIPAA, and GDPR obligations. Use unified tracing to debug slow steps, and autoscale backends to handle peaks without manual tuning. If you need to move workloads between regions or vendors, flip targets in the deployment layer and keep the same agent contract. With 24/7 support and SLAs, you can treat agents as dependable services—planned, measured, and continuously improved.
Developer
Free
Requests per month - 50K
Users - 3
Al Gateway
<ul>
<li>Universal AΡΙ
RBAC on models
Virtual models
Self-hosted models
Playground
Observability
<ul>
<li>Logs
Traces
Custom Metadata
Custom pricing / model
Cost per team/user/model/application
Metadata filtering
MCP Gateway
<ul>
<li>MCP Servers - Register up to 5
Tool calls per month - 50K
RBAC on MCPS
Metrics
Logs
Support for advanced authentication
Self-hosted MCPs
Prompt Management
<ul>
<li>Number of saved prompts - Up to 10
Versioning & Variables
Security & Authentication
<ul>
<li>SOC2
GDPR, HIPAA Compliance Certificates
Deployment modes
<ul>
<li>SaaS
Deployment customization
<ul>
<li>Gitops (Infrastructure as code)
Support
<ul>
<li>Support - Community Support
Pro
$499.00 per month
Requests per month - 1M
Users - 10
Al Gateway
<ul>
<li>Universal AΡΙ
RBAC on models
Self-hosted models
Playground
Simple caching
Semantic Caching
Control Center
<ul>
<li>Weight-based Routing
Latency-based Routing
Priority-based Routing
Fallbacks - With advanced features
Budget limiting
Rate limiting
Observability
<ul>
<li>Logs
Traces
Custom Metadata
Custom pricing / model
Cost per team/user/model/application
Metadata filtering
MCP Gateway
<ul>
<li>MCP Servers - Register up to 25
Tool calls per month - 1M
RBAC on MCPS
Metrics
Logs
Support for advanced authentication
Self-hosted MCPs
Prompt Management
<ul>
<li>Number of saved prompts - Unlimited
Versioning & Variables
Guardrails
<ul>
<li>Partner Guardrails integration
Security & Authentication
<ul>
<li>Role based access control
SOC2
Deployment modes
<ul>
<li>SaaS
Deployment customization
<ul>
<li>Gitops (Infrastructure as code)
Support
<ul>
<li>Support - Production
SLA - Standard SLA
Enterprise
Custom
Requests per month - Custom 10M Plus per month
Users - Custom
Al Gateway
<ul>
<li>Universal AΡΙ
RBAC on models
Virtual models
Self-hosted models
Playground
Multiple gateway endpoints
Simple caching
Semantic Caching
Control Center
<ul>
<li>Weight-based Routing
Latency-based Routing
Priority-based Routing
Fallbacks - With advanced features
Budget limiting
Rate limiting
Observability
<ul>
<li>Logs - With custom retention
Traces - Export to custom storage buckets
Feedback on traces
Custom Metadata
Custom pricing / model
Cost per team/user/model/application
Metadata filtering
Alerts
Export to other monitoring platforms
MCP Gateway
<ul>
<li>MCP Servers - Custom
Tool calls per month - Custom
RBAC on MCPS
Virtual MCP Servers
Metrics - Comprehensive
Logs
Support for advanced authentication
Self-hosted MCPs
Prompt Management
<ul>
<li>Number of saved prompts - Unlimited
Versioning & Variables
Guardrails
<ul>
<li>Partner Guardrails integration
Custom Guardrail Hooks
Security & Authentication
<ul>
<li>Role based access control
SSO
SOC2
GDPR, HIPAA Compliance Certificates
Org management
Audit logs
Deployment modes
<ul>
<li>SaaS
VPC / On-prem
Air-gapped deployment
Deployment customization
<ul>
<li>Data Lake Export
Connect multiple storage bucket
Multiple gateway planes
Gitops (Infrastructure as code)
Support
<ul>
<li>Support - Priority Support,Dedicated Onboarding
SLA - Enterprise-Grade SLA
Comments