Hermes Agent now works with Portkey as a drop-in gateway

Point Hermes at Portkey's OpenAI-compatible endpoint and you get 3,500+ models, full request logging, cost tracking per session, budget limits, and MCP governance at the gateway layer.

One line in config.yaml. Nothing else changes.

Docs: https://portkey.ai/docs/integrations/libraries/hermes-agent

Register any agent, swap in your Portkey endpoint, and get a governed production stack immediately. No changes to your agent code required.

What's included:

• Governance and access control scoped per agent, team, and user, with budget and usage limits enforced everywhere

• Full observability across 40+ metrics with complete traces

• Automatic fallbacks and load balancing so provider outages don't become your problem

• 50+ security guardrails covering PII, PHI, and content moderation with no custom setup

• One auth model across all your agents, regardless of framework

The control plane sits separately from your agents. Enforce policy, track costs, and kill any agent in real time with no redeployment required.

Beta is open.

Webinar on April 24. Register here now!

Pi is the fastest, leanest coding agent out there. You can now run it through Portkey.

By running Pi through Portkey, you get:

• Full observability across usage, latency, and spend
• Budget and token limits so long agentic sessions
• Fallbacks, retries, and caching across 3000+ models
• Governance and access controls baked in

One config change to set up. Docs here.

We just shipped a new guide for using Portkey with Conductor, the macOS AI coding app built on top of Claude Code.

With Portkey, you can:

• plug into 1600+ models through Portkey’s gateway,

• add fallbacks, load balancing, caching and routing,

• set cost / token limits for long agentic coding sessions,
• get full request‑level logs and metrics for every session.

To implement Portkey with conductor, checkout the docs here

You can now point Portkey at credentials in AWS Secrets Manager, Azure Key Vault, or HashiCorp Vault instead of pasting keys in the product—using secret_mappings so values are fetched at runtime.

This means:

→ Sensitive material stays in vaults you already control and audit

→ Integrations and virtual keys reference secrets by mapping, not plaintext

→ Rotation and access policies live where your infra team expects them

Set it up for your org ->

Sometimes your guardrails have dependencies. You might want to:

→ Run a PII detection check first, then only run compliance validation on the sanitized output

→ Check the request format before running an expensive LLM-based content safety check

→ Validate HIPAA compliance before running clinical content checks — no point evaluating medical accuracy if the request already violates patient data regulations

With the new sequential flag, checks run one after another in the order you define, and if an earlier check fails, the rest don't execute.

See how you can set up sequential guardrails →

We have some exciting news to share: Portkey has officially closed our $15M Series A !!

Huge thanks to our investors at Elevation and Lightspeed, as well as our team and community for getting us here.

You have pushed us to be better, given us helpful feedback, and suggested the features that have made Portkey what it is today!

This funding will continue expanding our go-to-market presence, deepen our governance capabilities for agentic AI, and scale the gateway that enterprises rely on to run AI in production.

What’s more, we’re making our core enterprise gateway available for free so that all teams have access to governance and observability when they need it!!

You can now route requests with the same identifier consistently to the same target when load balancing.

This means:

→ Conversation context stays intact across multiple requests

→ A/B tests produce consistent, reliable results

→ Users get session-based routing for personalized experiences

See how to set up this config ->

Claude Sonnet 4.6 is Anthropic's new default model, delivering near-Opus level reasoning and a 1M token context window, all at mid-tier pricing. Built for agentic workflows, computer use, and long-context tasks.

With Portkey, bring Claude Sonnet 4.6 into production with:

✅ Unified access via the AI Gateway

✅ Consistent guardrails and policy enforcement

✅ Full observability across logs, latency, and spend

✅ Team-level budgets and rate-limits

GPT-5.3-Codex is OpenAI’s latest model, designed for agentic, long-running coding workflows, covering everything from multi-step refactors to debugging across large repositories.

With Portkey, you can bring GPT-5.3-Codex into production with:

✅ Unified access via the AI Gateway

✅ Consistent guardrails and policy enforcement

✅ Full observability across logs, latency, and spend

✅ Team-level budgets and rate-limit controls