TrueFoundry is a unified platform with an enterprise-grade AI Gateway - combining LLM, MCP, and Agent Gateway to securely manage, route, and govern AI workloads
San Francisco
- Tweets 415
- Following 57
- Followers 1,193
- Likes 337
Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna shipped yesterday.
All three are live in the TrueFoundry AI Gateway, pricing included. Same endpoint, same key, new model string.
claude-opus-5-5 · $4 / $20
gpt-6-sol · $2 / $10
gpt-6-luna · $0.10 / $0.50
(per 1M tokens, in/out)
Made with AI
2
1
2
163
How to enable them (about a minute), what to check before moving production traffic: truefoundry.com/blog/claude-…
59
Live from the TrueForge hackathon in Santa Clara. Teams are shipping agents right now - real builds, not slideware.
Live updates in this thread 👇
11
3
13
452
Can we eliminate all Beta testers ? This team built up Probe to answer this problem .
piped.video/shorts/5QcplqByq…
2
92
TrueFoundry retweeted
Shyam who works at @PaloAltoNtwks building agent for support optimization.
1
1
103
TrueFoundry retweeted
It’s amazing to see the creativity of the group building in the hackathon. 2 hours to submission! Let’s go building.
1
1
124
TrueFoundry retweeted
Derick Lee building agents to manage thousands of programs for tens of thousands of students. Remove operational complexity of program management through agents!
1
2
100
TrueFoundry retweeted
@Rikin_786 crushing it at the hackathon building Mediforge- a medical agent built with TrueForge!
3
3
176
Here’s one of our participants talking about taking cybersecurity to the future with TrueForge 🚀
1
2
123
TrueFoundry retweeted
@TrueFoundry has been named the recipient of @FrostSullivan's 2026 Global Transformational Innovation Leadership Recognition in the enterprise #AI control plane category, following a 12-month independent evaluation. The report cites TrueFoundry's split-plane architecture, which enforces #data residency as a hard constraint rather than routing around gaps in compliant infrastructure.
Read more below:
frost.com/news/press-release…
3
9
509
Sept 19. Santa Clara. One day to take your agent from demo to something that survives production.
The Agent Harness Hackathon is here-spots are closing.
Register here : luma.com/truefoundry-agent-h…
Engineers and founders: turn your agent idea into something real.
Join the @truefoundry Agent Harness Hackathon Sept. 19 at AI Global Campus in Santa Clara. Bring a use case and build beyond the prototype.
Register: luma.com/truefoundry-agent-h…
2
5
300
Production traffic is not uniform.
You get a few requests that need your best model, but most are simple questions and lookups you can solve with cheaper, faster models.
The most expensive mistake you can make today is sending every request to your strongest model.
You need routing. Period.
This is the simplest trick to improve the architecture of whatever you are building.
Please, don't implement routing yourself. You don't have to.
I'm currently working with TrueFoundry's Auto Routing. It reads each request, classifies it as simple, medium, or complex, and sends it to the model assigned to that tier.
You have two choices:
1. Send every request to the free heuristic classifier to score signals such as technical vocabulary, code, prompt length, and multi-step reasoning.
2. Send the request to an LLM classifier when its difficulty requires a more nuanced judgment.
The beauty of using routing is that nothing changes in your code. You still call a single endpoint model, but routing works behind the scenes to pair every request with the best possible model.
TrueFoundry ran several experiments with two different setups:
1. Send every request to Claude Opus
2. Send every request to a router with Haiku, Sonnet, and Opus
The first experiment ran 550 deterministically graded academic prompts through every setup. Auto Routing was 69% cheaper while retaining 98% of the baseline quality.
The second experiment ran three production-shaped workloads through every setup, using user chats, developer chats, and long agent tasks. Auto Routing was 80% cheaper.
Thanks to the TrueFoundry team for partnering with me on this post.
15
9
36
15,610
Three agents worth actually running, built in a weekend on the harness: autonomous incident response with human approval, a sandbox that audits untrusted skills and MCP servers by watching what they do, and a dependency agent that fixes breaking changes and verifies the repair.
Congrats Elijah, Nelson, @Bissu99 and Priyanshu 🏆
Announcing the Agent Harness Hackathon winners! 🏆
Here are the winners who stood out with their projects and made an impact:
🏆 Elijah Umana
Best Use of @truefoundry → NVIDIA DGX Spark
Built oncall, an autonomous incident-response agent that investigates production issues and handles remediation with human approval.
📱 Nelson Lai
Best Code Quality @QodoAI → Mac Mini
Built CapyGuard, a security-focused agent that tests untrusted skills and MCP servers by observing what they actually do.
🎧 @Bissu99 & Priyanshu K Solanki
Best Use of Bright Data → Apple AirPods
Built AgentRadar, an autonomous dependency-repair agent that detects breaking changes, fixes affected code, and verifies the repair.
Huge congratulations to all our winners, and thank you to everyone who built with us! 🚀
Ready to build something of your own? Explore the Bharat Builds Tour: wemakedevs.org/aws
3
4
16
1,360
TrueFoundry is officially an @msPartner! We have been selected for the exclusive, invite-only Microsoft for Startups Pegasus program for high-growth startups.
We’re excited to work more closely with Microsoft as we continue building the infrastructure enterprises need to connect, observe, and govern their AI workloads.
Find us on the Microsoft Marketplace!
Made with AI
2
6
9
391
Install to working agent in 8 minutes - on a model running on your own hardware.
one npx command. one SQLite file.
@dr_cintas built the whole thing and filmed it 👇
run agent harnesses on local models.
(no token costs, no model API keys, 100% open-source)
your agent runs on your machine. the model doesn't. every prompt, every file, and every secret leaves your laptop before the agent does anything with it.
this new open-source fixes that. it's an agent harness that takes any OpenAI-compatible endpoint, which means Ollama, LM Studio, or vLLM running on your own hardware. no provider key, no per-token bill.
setup is one command. local mode is a single process with a SQLite file, so there's nothing else to install and no database to stand up.
then you configure four things once: models, MCP servers, skills, and a sandbox (Daytona today; local sandbox execution is on the roadmap). every agent you build after that picks from what you connected.
everyday work it covers:
→ read and analyze private documents
→ query internal APIs over MCP
→ review code and logs
→ search and organize local files
→ run multi-step workflows from one prompt
MIT licensed. fork it, self host it, ship it inside a commercial product.
𝗻𝗽𝘅 @𝘁𝗿𝘂𝗲𝗳𝗼𝘂𝗻𝗱𝗿𝘆/𝘁𝗿𝘂𝗲𝗳𝗼𝗿𝗴𝗲@𝗹𝗮𝘁𝗲𝘀𝘁
I built a complete agent with it in 8 minutes. full walkthrough below.
1
9
411