Turns products into CLIs and agents into customers
74107451575K171
Tags
AI agentSeedB2D (developers)Product launch500K-1MExplainerGlobalFounder-led
TLVC Rating
3.00
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy
Community Rating
No ratings yet
to rate this launch.
Okibi generates a command line tool from a company's product so that AI agents can operate it directly, the same way a human user would run commands in a terminal. The pitch, aimed at API and SaaS builders staring down a wave of agent traffic, is that a handful of MCP tools can't cover the surface area of a real product and letting an agent build every API call from scratch tends to produce broken calls, while a purpose-built CLI ships commands that already work . This YC S25 launch matters now because agents are starting to sign up, transact, and integrate on behalf of users, and most SaaS products have no dedicated surface for them.
Under the hood, Okibi researches a repo alongside competitor CLIs, MCP servers, and SDKs, drafts a command set that the founder can steer, then builds and verifies each command in a cloud sandbox with structured JSON output and dry-run behavior on writes . Each generated CLI ships with a skill file so an installing agent already knows how to use the product, and Okibi maintains the CLI as the underlying product changes . A companion piece called Okibi Identity tackles the sign-up wall, letting an agent create an account on its user's behalf and carry short-lived signed proof of who it is acting for on every request, with one identity working across every Okibi CLI .
Okibi is the second YC company from Mahyad Ghassemi and Saurav Mitra, who founded it in 2025 and run a two-person team out of San Francisco . Mahyad previously built SigmaOS, a web browser designed for work , and the team's earlier Okibi product was a natural-language agent builder before the company refocused on the CLI and identity layer that lets other founders turn their own software into something agents can actually buy from.
Comments (15)
Sign in to join the discussion.
P
Priya Ramanathan24d ago
Cute framing but who pays when the agent uses the CLI? If OpenAI is the customer, margins get squeezed to zero the moment they ship their own.
B
Bekzod Rakhimov24d ago
Retention question: once an agent framework standardizes on MCP or similar, why does the CLI layer survive? Feels like a bridge asset.
T
Tunde Okafor24d ago
Genuinely curious about auth flows and rate limits per agent identity. If I spin up 400 agents on the same account do you throttle me or bill me into oblivion?
C
Chinelo Adeyemi24d ago
Also, do agents get their own API keys with scoped permissions or is it one key to rule them all? Security posture matters if this ends up in prod.
M
maxi24d ago
This is one of the last SaaS products anyone will build a human UI for. In 18 months every dashboard will be a fallback surface for when the agent breaks.
L
Lin Haoyu24d ago
The demo skips the boring part: how do agents discover which CLI to install? Distribution is the whole game here and nobody talks about it.
R
Rafa Delgado24d ago
The wordmark on that landing page has kerning between the k and the i that made me physically flinch. Fix that before agents start judging you too.
H
Hira Suleiman24d ago
ok wait, turning every product into a CLI is actually the most based positioning I've seen this month. MCP kids in shambles.
D
Dr. Yaro Zelenko24d ago
Reminds me of the tool-use benchmarks from last year where structured CLI interfaces outperformed REST wrappers by a decent margin. Would love to see your eval methodology.
Y
Yui N.24d ago
The launch video's cold open is 4 seconds too long and the zoom on the terminal at 0:23 saved the whole thing. Whoever edited this, hire them again.
S
Sana Iqbal24d ago
The jump cut between the human clicking and the agent typing is the entire pitch in one second. That's the frame you should've made the thumbnail.
D
Devansh K.24d ago
Impressions look healthy but the like-to-bookmark ratio on this tweet is weird. People are saving it to try later, which usually means the pitch didn't fully land in one read.
M
Marta Kowalczyk24d ago
How many people built this? If it's more than three I'm going to be disappointed given how narrow the wedge is.
G
Grigor Petrosyan24d ago
hot take: CLIs are just the training wheels. In a year the interface layer collapses and agents talk to intents directly.
M
Mireille Vasquez24d ago
Tweet copy is clean but the second line reads like a Notion doc heading. 'most apps were designed for people, not for agents' could have been the whole tweet.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Okibi AI
Turns products into CLIs and agents into customers
Turns products into CLIs and agents into customers
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Okibi generates a command line tool from a company's product so that AI agents can operate it directly, the same way a human user would run commands in a terminal. The pitch, aimed at API and SaaS builders staring down a wave of agent traffic, is that a handful of MCP tools can't cover the surface area of a real product and letting an agent build every API call from scratch tends to produce broken calls, while a purpose-built CLI ships commands that already work . This YC S25 launch matters now because agents are starting to sign up, transact, and integrate on behalf of users, and most SaaS products have no dedicated surface for them. Under the hood, Okibi researches a repo alongside competitor CLIs, MCP servers, and SDKs, drafts a command set that the founder can steer, then builds and verifies each command in a cloud sandbox with structured JSON output and dry-run behavior on writes . Each generated CLI ships with a skill file so an installing agent already knows how to use the product, and Okibi maintains the CLI as the underlying product changes . A companion piece called Okibi Identity tackles the sign-up wall, letting an agent create an account on its user's behalf and carry short-lived signed proof of who it is acting for on every request, with one identity working across every Okibi CLI . Okibi is the second YC company from Mahyad Ghassemi and Saurav Mitra, who founded it in 2025 and run a two-person team out of San Francisco . Mahyad previously built SigmaOS, a web browser designed for work , and the team's earlier Okibi product was a natural-language agent builder before the company refocused on the CLI and identity layer that lets other founders turn their own software into something agents can actually buy from.
Comments (15)
Cute framing but who pays when the agent uses the CLI? If OpenAI is the customer, margins get squeezed to zero the moment they ship their own.
Retention question: once an agent framework standardizes on MCP or similar, why does the CLI layer survive? Feels like a bridge asset.
Genuinely curious about auth flows and rate limits per agent identity. If I spin up 400 agents on the same account do you throttle me or bill me into oblivion?
Also, do agents get their own API keys with scoped permissions or is it one key to rule them all? Security posture matters if this ends up in prod.
This is one of the last SaaS products anyone will build a human UI for. In 18 months every dashboard will be a fallback surface for when the agent breaks.
The demo skips the boring part: how do agents discover which CLI to install? Distribution is the whole game here and nobody talks about it.
The wordmark on that landing page has kerning between the k and the i that made me physically flinch. Fix that before agents start judging you too.
ok wait, turning every product into a CLI is actually the most based positioning I've seen this month. MCP kids in shambles.
Reminds me of the tool-use benchmarks from last year where structured CLI interfaces outperformed REST wrappers by a decent margin. Would love to see your eval methodology.
The launch video's cold open is 4 seconds too long and the zoom on the terminal at 0:23 saved the whole thing. Whoever edited this, hire them again.
The jump cut between the human clicking and the agent typing is the entire pitch in one second. That's the frame you should've made the thumbnail.
Impressions look healthy but the like-to-bookmark ratio on this tweet is weird. People are saving it to try later, which usually means the pitch didn't fully land in one read.
How many people built this? If it's more than three I'm going to be disappointed given how narrow the wedge is.
hot take: CLIs are just the training wheels. In a year the interface layer collapses and agents talk to intents directly.
Tweet copy is clean but the second line reads like a Notion doc heading. 'most apps were designed for people, not for agents' could have been the whole tweet.