How Work Gets Done
At LiteLLM, the engineering calendar moves faster than the model calendar it tracks. The team commits to day-zero releases for new models from OpenAI, Anthropic, Google, and the 140-plus providers that route through its gateway; one engineering testimonial on the company's site says they deliver "the latest LLM models to our users, usually within a day of them being released."
The commit count tells part of the story. The public repository carries 46,232 commits since its Y Combinator W23 launch, sustained across a small, hybrid team, and that volume is how the gateway stays current with a provider landscape that adds endpoints, retires old ones, and ships reasoning variants every few weeks.
Open-source development is the workflow, not a side project. LiteLLM sits under an MIT license, and its GitHub repository lists 1,005-plus contributors and 53,000-plus stars. Releases land in a public changelog, security disclosures ship next to fixes, and pull requests from unfamiliar names land in the same review queue as internal work.
Authority inside the codebase is concentrated where the throughput demands it. The LiteLLM Rust AI Gateway benchmark figures, 0.66 ms at p99, 2,800-plus requests per second at roughly 21% CPU, about 22 MB at rest, were measured on a Rust core that sits in front of the Python SDK that defined the original product. Decisions about what stays in the SDK and what moves to the Rust gateway reflect a small, technically senior team picking its fights.
Customer signal flows into the same loop. Spend tracking, virtual keys, budget caps, and per-team rate limits are gateway features, and the same telemetry that shows a customer where their dollars go gives engineers feedback on which providers and prompts are slow, expensive, or flaky. NVIDIA's team, cited on LiteLLM's site, says the gateway gives them "a single, consistent way to access more than 100 AI model endpoints," and one enterprise customer reported cost drops of up to 56% on coding workloads after switching on Auto Routing.
The team's hiring shape reinforces the rhythm. Recent postings on Zero G Talent's board put it at 15 salaried roles with a median around $230,000 and a band running $96,000 to $276,000. The shape of that board (one head of engineering, three staff or senior staff engineers, a dedicated Rust seat, a security hire, product engineers carrying customer-facing surface area) mirrors how decisions actually move. A founder or the head of engineering sets the model-support calendar; staff and principal engineers draw the line between Rust and Python; the table below shows the salary bands that anchor the technical spine. The pace is fast, the org is flat, and the person closest to the code usually has the final word.
| Role | Salary Band |
|---|---|
| Head of Engineering | $250,000–$300,000 |
| Staff Engineer | $240,000–$270,000 |
| Rust Engineer | $220,000–$280,000 |
| Product Engineer | $200,000–$260,000 |
All based in San Francisco with some remote US seats for security and backend work.
What the Company Actually Stands For
LiteLLM does not publish a glossy culture deck, and that absence is the closest thing to a value statement the team makes. The clearest articulation lives on litellm.ai, where the team describes itself as "the fastest, litest AI Gateway" with a "Rust core," phrasing that fuses engineering taste to operating identity. Berri AI Incorporated, the entity behind the open-source library, went through Y Combinator's W23 batch, and the priorities that mark a YC-stage company (fast iteration, narrow scope, founder-led decisions) show up directly in how LiteLLM talks about itself.
Interoperability is the first principle, treated as a moral stance rather than a feature. LiteLLM exists because, in the founders' words on the project README, "managing LLM calls across providers gets complicated fast — different SDKs, auth patterns, request formats, and error types for every model." The response is a single OpenAI-compatible interface that now spans 140-plus providers, roughly 1,892 unique models, and endpoints for chat, responses, embeddings, images, audio, batches, rerank, and the agent and MCP surfaces that have come online since. "Drop-in OpenAI compatibility" and "no lock-in" repeat across the README, the docs, and the marketing site. The MIT license is treated as both a technical and a philosophical commitment. The self-hosted gateway with 240 million-plus Docker pulls behind it and the explicit "free forever" tier exist because the team has chosen to remove every commercial barrier between a developer and the code.
Performance is the second principle, served as a public proof point. LiteLLM publishes its own benchmarks rather than leaving them to inference. According to the site, it claims 8ms P95 latency at 1,000 requests per second, roughly 0.66 ms overhead at p99, throughput above 2,800 requests per second at around 21% CPU, and "about 4.5× more requests per dollar" than the next-fastest gateway. The comparison names competitors directly, Portkey at 2.29 ms and Bifrost at 4.54 ms, and ties the numbers to an in-house tool called AI Gateway Bench, which the company calls "an open standard." Leading with measured, third-party-comparable numbers rather than adjectives is itself a value statement about how a product should earn trust.
Underneath those two sits a quieter principle of transparency under pressure. The site says, "When there's an issue, we publish the disclosure and the fix. You can read every one," a sentence that reads as both a security practice and a culture tell. Combined with the open-source posture (53,000-plus GitHub stars, 11.1k forks, 46,232 commits, 1,005-plus contributors), the operating principles add up to a team that ships in public, argues for its choices with numbers, and treats the absence of lock-in as a feature.
What the Hiring Bar Selects For
LiteLLM's open job board tells you almost everything the company wants in a hire without a recruiter having to say it. Every posted role clusters in San Francisco, with a single remote exception for a senior security engineer. Every role lists a base salary band of roughly $200,000 to $300,000. And every role, from Head of Engineering down to Product Engineer, asks for a candidate who can ship code, defend a high-traffic open-source package, and operate one pace ahead of attackers. The hiring bar is narrow in geography and wide in expectation.
The supply-chain incident of March 24, 2026, when attackers tracked as TeamPCP uploaded compromised versions 1.82.7 and 1.82.8 to PyPI, exposed how much a successful LiteLLM hire has to carry. The malicious releases stayed live for roughly 40 minutes; in that window, CloudSEK's reconstructed dataset tied the campaign to 2,500-plus organizations and roughly 434,000 potentially affected CI/CD pipelines, with high-confidence matches at Amazon Web Services, Cisco, ServiceNow, Siemens, John Deere, London Stock Exchange Group, FedEx, MediaTek, Volkswagen, Deloitte, Kroger, Thales, X Corp, Zscaler, Orange, Fortum, and Krungthai Bank. CloudSEK's writeup also names AI gateways and agent runtimes as the new "strategic targets" because they sit at the junction of cloud, code, and model credentials.
Customer proximity is the cultural filter. The attack pushed LiteLLM to switch compliance vendors, moving from a startup called Delve to Vanta for certifications, a vendor decision made inside a 72-hour incident window. That kind of move tells you what the company rewards in people: a bias toward shipping a fix over debating one. Headcount on the board is small and concentrated in product engineering.
Technical depth is the floor. Across 15 salaried roles, recent postings on Zero G Talent's board show the median sits near $230,000, with the full range spanning about $96,000 to $276,000. LiteLLM is paying against the same band that OpenAI and Anthropic use for senior ICs, even though LiteLLM runs as a small open-source gateway.
Is There Anything to Read About Working There?
There is essentially no public employee review of LiteLLM available. The research digest for this section pulled up financial forum threads about leveraged ETFs, an Excel comment-image tutorial, and a checkbox insertion guide, nothing from Glassdoor, Blind, Comparably, Levels.fyi, or any other workplace-review platform. That absence is the most important fact a job seeker can act on.
For a Series A-stage developer-tools company with 15 salaried roles currently posted on Zero G Talent's board, the lack of a Glassdoor page, a Blind thread, or even a handful of LinkedIn exit-announcements means prospective hires cannot triangulate the day-to-day the way they could at a bigger employer.
This creates a real risk for candidates. Founders at small developer-tools companies can describe a culture of technical rigor and customer focus that the team genuinely lives, but the absence of corroborating reviews means the reader has to weigh that self-description without a counterweight.
For now, the practical move is to ask directly in the interview loop. Ask the founders and the hiring managers what current and former employees have told them in private, and what the attrition pattern has looked like over the past twelve months.
Who Thrives and Who Burns Out
At a small hybrid company shipping open-source LLM infrastructure, the line between thriving and burning out is drawn less by raw talent than by tolerance for ambiguity, ownership, and the cadence of customer-driven iteration. The roles on the Zero G Talent board tell you something about who the company is willing to pay to keep, but they don't tell you who lasts. The roles themselves do. Head of Engineering, plus Staff, Senior Backend, Senior Security, Rust, and Product Engineers clustered around San Francisco or remote-US: every listing is a senior, high-autonomy title. There are no junior seats, no apprenticeship track, no "learn on the job" postings, and that alone filters out a category of candidate who would otherwise have applied.
The people who tend to do well in this kind of shop share a recognizable profile. They are comfortable owning a feature from user complaint through pull request through Discord reply, because the customer feedback loop at a small open-source project is tight, public, and immediate. Zero G Talent found the "Product Engineer" and "Senior Backend Engineer" listings both cluster at $200,000–$260,000, and that overlap is a signal: LiteLLM is not separating implementation from product judgment the way a larger org would.
Geography matters here in a way that pure remote-first rhetoric misses. Five of the seven current board postings are San Francisco–based or San Francisco–preferred, with remote-US as a fallback for a couple of security and Rust roles. The implication is that the founding team is in the room, literally, for the bulk of the work.
The burnout profile is the inverse of the thriving profile, and the gap between them is narrow. The same autonomy that lets a Staff Engineer ship a routing fix in a single afternoon also means no one is paging them when they should sleep. At a hybrid team of roughly a dozen engineers running fast-paced release cycles, there is no release manager, no QA buffer, no on-call rotation with eight people on it.
Two patterns deserve naming. First, the salary band on the board is wide enough ($96,000 at the floor, $276,000 at the top, with most roles paying between $200,000 and $280,000) that compensation alone will not retain someone who is mismatched on working style. Second, large enterprises on LiteLLM's customer base carry exactly the procurement complexity that bleeds into engineering prioritization.
Working in frontier tech? Zero G Talent tracks the openings: see every open LiteLLM role, browse frontier tech jobs, the companies hiring, and the people building the field.