TrueFoundry

We're running a TrueForge track at Stranger builds: build a self-improving agent 🔁 an agent that gets better every run beats one that's perfect on day one. show us yours. See you Saturday!

zero users is a bug. you have 7 days 🎃 stranger builds starts saturday: SF 🇺🇸 Melbourne 🇦🇺 Bangalore 🇮🇳 online 🌍 fix a problem, get your first users. $10K in prizes and credits thank you @Fetch_ai @Base44 @VonageDev, @truefoundry (TrueForge) and @ASILab for betting on nerds 🤓 your friend with 0 users will love this, send it over! 👀

Agent Identity is live on TrueFoundry 🔐 Every AI agent gets its own verifiable identity, enforced at the gateway that already runs your LLM and MCP traffic. No shared keys. No auth code in your agents.

Now you can decide who can call each agent and what each agent can call. Works with agents on Bedrock AgentCore, Vertex AI, LangGraph and your own services.

Multi-agent chains lose the user by hop three. Agent Identity lets you choose how identity flows at each hop: pass-through, client credentials, on-behalf-of token exchange (Entra + Okta) or AWS SigV4.

GPT-6.1 Terra is live on TrueFoundry AI Gateway. $1 / $5 per 1M tokens, half of GPT-6.1 Sol. 1.05M-token context, 128K output. Model string: openai/gpt-6.1-terra. Same endpoint, same keys. Read more :

How to switch: tick the model in Models Selection, set Access Control, swap the model string. Nothing else changes. OpenRouter and AWS Bedrock (global and US profiles) are in the catalog now; OpenAI direct and Azure entries will follow.

Where it fits: Terra for the broad middle of your traffic, Sol for the hard tail, one virtual model with Auto Routing in front. Pricing is public at

₹3L in prizes + PPIs, up for grabs. 🏆 The TrueFoundry Forge-a-thon is live at @Techfest_IITB .Build, deploy and scale AI agents with TrueForge. Open to students and working professionals. Grand Finale: Dec 2026. Registration link in comments .

Register on this link :

RT @IStartupTimes: The AI Founders 100 — India’s Rising Leaders Building the Future with AI Read More:

TrueFoundry now integrates with @confident_ai. Teams routing LLM traffic through the TrueFoundry AI Gateway can now send every request to Confident AI as an OpenTelemetry trace and score it with online evaluations: answer relevancy, faithfulness, toxicity, and PII leakage, all running on live production traffic, with zero tracing code in your applications. Evaluation runs asynchronously, off the request path, so it adds no latency to user-facing calls. Combined with TrueFoundry’s unified access to 1,000+ LLMs, rate limiting, cost tracking, and governance, teams get one place to route every model call and see how good each answer actually was, with quality scores sitting right next to latency and cost. One gateway. Every request, evaluated. Thanks to the @confident_ai for the collaboration!

The Decisions API is live on TrueFoundry AI Gateway: ask typed yes/no, choice and score questions, get probabilities back. 7 providers, 2 routes: /v1/decisions and /v1/systemone. Same keys; switch provider by changing the model string. Read more:

3/ Limits to plan for: no streaming. Image input works only with OpenAI models on /v1/decisions. Self-hosted vLLM answers choice questions only.

4/ Tracing, cost tracking (input tokens only), budgets and rate limits apply as for any model, and a virtual model can fail over across OpenAI and System One targets. Docs:

Using TrueForge with Jev produces interesting results, to say the least !

Here's OperatorOS, the first-place build from our recent #AgentsThatAct hackathon, in action 🎬 Neighborhood store owners in India spend hours every day reconciling cash, restocking, chasing customer credit and running promotions, across tools that mostly only speak English. Our first-place winners built a multilingual agent that runs that whole process end-to-end, by voice. The entire workflow was built on the TrueForge Agent Harness and TrueFoundry AI Gateway.

GPT-6.1 Sol is live on TrueFoundry AI Gateway. $2 / $10 per 1M tokens, one-fifth of GPT-6 Astra's list price, with cache reads at $0.10 (half of GPT-6 Sol). Model string: openai/gpt-6.1-sol. Same endpoint, same keys.

How to switch: tick the model in Models Selection, set Access Control, swap the model string. Nothing else changes. Read more here :

Where it fits: GPT-6 Luna for simple requests, GPT-6.1 Sol for medium, GPT-6 Astra for the hard tail, one virtual model with Auto Routing in front. Pricing is public at