Portkey
@PortkeyAI
Followers
2K
Following
1K
Media
406
Statuses
1K
AI (Gateway, Guardrails, Governance)∙ Processing 50 Billion+ LLM tokens every day. Backed by @lightspeedvp | 8,100+ Stars
SF, BLR
Joined March 2023
We just launched Model Catalog!!. If your org has multiple teams building with LLMs, this is the control layer you’ve probably been missing. 🧵.
2
5
8
Not every team wants to log everything. Some need full visibility into every LLM request. Others have strict compliance rules and want to keep things minimal. Portkey now gives org owners control over what gets logged, from full request details to just high-level metrics.
0
0
3
happening tomorrow!!. register if you haven't already: @mlopscommunity @retrovrv.
MCP is here. Or is it? 👀. @retrovrv will be looking at where things actually stand at the Agents in Production conference (virtual) on July 17. He’ll be joining peers from Microsoft, Palona, HP, Baseten, Stanford, and more. @mlopscommunity
0
4
7
Portkey can help you with that. Now you can plug Portkey in Langflow and get:. ✔ Support for 1600+ LLMs.✔ Full observability across every call.✔ Budget limits + RBAC.✔ Guardrails for security and compliance. Full guide →
0
0
2
Building the workflow is just the first step. You also need to:.– Track usage in real time.– Keep spends under control.– Set access control.– Know exactly what went wrong, when it does.
1
0
1
Run Kimi K2 in production with:. ✔ Smart routing + retries .✔ Guardrails for safe usage .✔ Full observability (logs, latency, cost) .✔ Budget + rate limits per team.
1
0
1
Kimi K2 is now live on Portkey 🚀.
🚀 Hello, Kimi K2! Open-Source Agentic Model!.🔹 1T total / 32B active MoE model.🔹 SOTA on SWE Bench Verified, Tau2 & AceBench among open models.🔹Strong in coding and agentic tasks.🐤 Multimodal & thought-mode not supported for now. With Kimi K2, advanced agentic intelligence
1
1
4
RT @TheNoahHein: Every serious AI app needs an API gateway. I have heard @swyx say this over a dozen different times. So This upcoming Mon….
0
4
0
when we were cutting cake to mark one year of Portkey, Stripe pinged with our biggest customer payment yet. timing so good, it felt scripted xD.
0
4
14
⚡ Day 0 support for Grok 4!. You can now start using @xai's Grok 4 via Portkey’s AI gateway. Grok 4 currently tops the Artificial Analysis Intelligence Index and outperforms others in coding and math tasks. With Portkey, bring Grok 4 into production with:. ✅ Smart routing,
2
1
7
Portkey now supports circuit breaker configuration. Because at scale, slow is the new downtime. Read more here ->
0
0
2
A circuit breaker watches for failure rates and rising response times. So, when a target crosses the threshold, it stops sending traffic there and routes it to healthier targets.
1
0
2
What you need instead is a proactive pattern — one that stops retrying operations likely to fail. That’s what circuit breakers are for.
1
0
2
Slow targets trigger retry storms →.Retries increase system load →.Fallbacks get delayed →.Latency rises for everyone. This is how degraded performance turns into cascading failure.
1
0
2
Retries and fallbacks are built for short, recoverable issues. They work, but only if the problem resolves quickly. When failures last longer, you need something better.
1
0
2
Transient slowdowns, 5xxs, and regional blips happen more often than you think. especially at scale.And retries alone won’t save you 🧵
1
0
4
If Claude Code is part of your stack, Portkey helps you run it beyond experiments. See how ->
0
0
3
With Portkey, you can:. ✅ Set budget limits .✅ Track usage by user, project, or team.✅ Enforce RBAC + approve model access.✅ Add observability to every call.✅ Handle retries, fallbacks, and guardrails at the gateway level.
1
0
2
– We keep hitting usage limits at the worst time — how do we prevent that?.– Can we track usage by user or team?.– Can we control who gets access?.– Can we choose which models teams can access?.
1
0
2