Omnara is an open-source API for building and running production-grade AI agents, aimed at developers who want to ship agents without stitching together their own infrastructure. It positions itself as the open-source alternative to Claude Managed Agents , handling durable execution, machines, tools, secrets, approvals, and streaming out of the box while remaining model-agnostic and self-hostable . The pitch to teams is straightforward: write the agent logic, and let Omnara take care of hosting, state, sandboxes, and permissions.
The launch matters because managed agent platforms from model vendors are quickly becoming the default path to production, and Omnara is offering an open, portable option for teams that do not want to lock into a single provider. The project ships with a Docker Compose setup that runs locally with a single command , and for teams that prefer not to self-host, Omnara offers hosted models through OpenRouter and hosted sandboxes through Blaxel, with $1,000 in credits available to the first 50 teams building on the platform .
Omnara is a Y Combinator S25 company founded in 2025 by Kartik Sarangmath and Ishaan Sehgal, with four employees based in San Francisco . Sehgal and Sarangmath are childhood friends with backgrounds in AI and AI infrastructure at companies including Meta and Microsoft , and Sehgal previously worked on the YC-backed startup Windsor and at Roblox before starting Omnara.
Comments (13)
Sign in to join the discussion.
P
Priya K.Sep 2, 2026
the API for production is doing a lot of heavy lifting there. production of what? potatoes? agents? podcasts? add one noun and you're golden.
T
Tomas LindqvistSep 2, 2026
durable state, sandboxes, secrets, permissions, hosting. that's five hard problems in one sentence and I need to see the reliability numbers before I believe any of them.
D
dev.kSep 2, 2026
the demo cut at 0:14 was suspicious. did the agent actually finish the task or did we jump to a pre-baked terminal?
Z
Zainab O.Sep 2, 2026
ok wait, open-source alternative to Claude Managed Agents is the sharpest positioning I've read this month. that's a full go-to-market in one line.
M
Maeve C.Sep 2, 2026
before anyone ships this into a regulated env, where does the durable state actually live and can I pin it to a region? asking for every enterprise buyer ever.
H
hiroakiSep 2, 2026
the thumbnail is fine but the title card font is fighting the code snippet underneath. two monospaced fonts walked into a bar and neither of them left.
F
Farah I.Sep 2, 2026
the sandboxes/permissions/secrets diagram is genuinely clean though. someone on that team knows their way around a whiteboard.
Y
Yusuf A.Sep 2, 2026
hot take: managed agent platforms are the new vector db wave. everyone launches one, three survive, the rest quietly pivot to observability.
B
Benoît Z.Sep 2, 2026
counterpoint on the pivot theory: infra for agents is the one bet that actually gets more valuable the more the model layer commoditizes.
L
Linh T.Sep 2, 2026
how big is the team on this? because if it's under six people I need to sit down and reconsider my life choices.
C
Chidera N.Sep 2, 2026
genuine question from someone who just learned what an agent is: if my agent crashes mid-task, does Omnara resume it or do I get charged for the failed run?
S
Svetlana R.Sep 2, 2026
the tweet buries the lede. "open-source alternative to Claude Managed Agents" should be the first line and the video's first frame, not the third clause.
R
Raoul M.Sep 2, 2026
you'd have doubled retention if the video opened on the sandbox spinning up instead of the logo animation. nobody stays for the logo, ishaan.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Omnara AI
the API for production-grade agents
the API for production-grade agents
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Omnara is an open-source API for building and running production-grade AI agents, aimed at developers who want to ship agents without stitching together their own infrastructure. It positions itself as the open-source alternative to Claude Managed Agents , handling durable execution, machines, tools, secrets, approvals, and streaming out of the box while remaining model-agnostic and self-hostable . The pitch to teams is straightforward: write the agent logic, and let Omnara take care of hosting, state, sandboxes, and permissions. The launch matters because managed agent platforms from model vendors are quickly becoming the default path to production, and Omnara is offering an open, portable option for teams that do not want to lock into a single provider. The project ships with a Docker Compose setup that runs locally with a single command , and for teams that prefer not to self-host, Omnara offers hosted models through OpenRouter and hosted sandboxes through Blaxel, with $1,000 in credits available to the first 50 teams building on the platform . Omnara is a Y Combinator S25 company founded in 2025 by Kartik Sarangmath and Ishaan Sehgal, with four employees based in San Francisco . Sehgal and Sarangmath are childhood friends with backgrounds in AI and AI infrastructure at companies including Meta and Microsoft , and Sehgal previously worked on the YC-backed startup Windsor and at Roblox before starting Omnara.
Comments (13)
the API for production is doing a lot of heavy lifting there. production of what? potatoes? agents? podcasts? add one noun and you're golden.
durable state, sandboxes, secrets, permissions, hosting. that's five hard problems in one sentence and I need to see the reliability numbers before I believe any of them.
the demo cut at 0:14 was suspicious. did the agent actually finish the task or did we jump to a pre-baked terminal?
ok wait, open-source alternative to Claude Managed Agents is the sharpest positioning I've read this month. that's a full go-to-market in one line.
before anyone ships this into a regulated env, where does the durable state actually live and can I pin it to a region? asking for every enterprise buyer ever.
the thumbnail is fine but the title card font is fighting the code snippet underneath. two monospaced fonts walked into a bar and neither of them left.
the sandboxes/permissions/secrets diagram is genuinely clean though. someone on that team knows their way around a whiteboard.
hot take: managed agent platforms are the new vector db wave. everyone launches one, three survive, the rest quietly pivot to observability.
counterpoint on the pivot theory: infra for agents is the one bet that actually gets more valuable the more the model layer commoditizes.
how big is the team on this? because if it's under six people I need to sit down and reconsider my life choices.
genuine question from someone who just learned what an agent is: if my agent crashes mid-task, does Omnara resume it or do I get charged for the failed run?
the tweet buries the lede. "open-source alternative to Claude Managed Agents" should be the first line and the video's first frame, not the third clause.
you'd have doubled retention if the video opened on the sandbox spinning up instead of the logo animation. nobody stays for the logo, ishaan.