SAASPOCALYPSEverdict #LITELLM-AAB1
scanned 2026.07.22 · 16:47
subject of investigation

litellm.ai

OpenAI-format LLM gateway
verdictSOFT
wedge score
84
/100
wedge thesis

the door is distribution and positioning: they're a protocol wrapper (OpenAI-format) with no proprietary data or unique integrations, so distribution and clear developer familiarity are the weakest defensible surfaces.

wide-open walls — wedgeable·ship in 2 weeks·run for $27.00 + usage
the doordata
wedge

where the walls are.

methodology →
the door

no proprietary corpus — they're running on off-the-shelf data.

capital
1.0/10
investment the incumbent had to make
why this scorehigh confidenceMinimal non-software spend or specialized capital; primarily cloud hosting and edge functions which are inexpensive...

Minimal non-software spend or specialized capital; primarily cloud hosting and edge functions which are inexpensive and replaceable.

  • Estimated competing cost lists only cheap hosting and edge exec (~$27)
  • Wedge is developer-focused protocol wrapper with no proprietary infra
technical
3.0/10
depth of the underlying engineering
why this scoremedium confidenceSome engineering complexity (streaming, load-balancing, provider mapping) but generally standard software challenges...

Some engineering complexity (streaming, load-balancing, provider mapping) but generally standard software challenges without deep technical secrets.

  • Challenges note streaming and provider credential security as hard
  • Product maps OpenAI-format to many provider APIs (engineering glue)
network
1.0/10
users compound users
why this scorehigh confidenceNo marketplace, UGC, social graph, or multi-sided liquidity; distribution is the primary moat, not network effects.

No marketplace, UGC, social graph, or multi-sided liquidity; distribution is the primary moat, not network effects.

  • Wedge thesis explicitly cites no network effect
  • Take_sub states no proprietary corpus or network effect
switching
2.0/10
stickiness of customer data + workflow
why this scoremedium confidenceUsers can likely switch easily since data is requests/passthroughs and keys; some migration friction for integrations...

Users can likely switch easily since data is requests/passthroughs and keys; some migration friction for integrations but not lock-in.

  • Service standardizes OpenAI-format but competitors can replicate mapping
  • Detected stack uses common components (Supabase, Cloudflare) enabling easy migration
datadoor
0.0/10
proprietary data accumulates over time
why this scorehigh confidenceNo proprietary training corpus, behavioral flywheel, or accumulated non-exportable dataset mentioned.

No proprietary training corpus, behavioral flywheel, or accumulated non-exportable dataset mentioned.

  • Take and take_sub state no proprietary data
  • Report lists only routing/auth and usage logs, not proprietary model data
regulatory
1.0/10
real licenses, not SOC 2 theater
why this scorehigh confidenceNo regulated duties or licenses indicated; SOC2-level concerns noted but not binding regulations like HIPAA or money...

No regulated duties or licenses indicated; SOC2-level concerns noted but not binding regulations like HIPAA or money transmission.

  • Challenges mention security and credential safety but not HIPAA/FINRA/KYC
  • Wedge and stack description include common cloud services without regulated compliance obligations
take

the blunt take.

This is a thin wrapper that standardizes many LLMs behind the OpenAI API shape — valuable, but not defensible once someone bundles it into an existing dev tool or OSS gateway; the play is developer-focused distribution, not technical one.

LiteLLM manages auth, load-balancing and spend across 100+ LLMs in OpenAI format — that's largely engineering glue and docs, not a proprietary corpus or network effect, so the fastest wedge is offering an easier dev experience, open-source SDKs, or one-click self-hosts.

cost

cost of competing.

what they charge
unknown — productized gateway pricing likely
tbd
/ plan/mo
Homepage shows product but no public pricing — competitors charge monthly for managed gateways
annual:tbd
what running yours costs
01 · Vercel (hobby) / static hosting$0.00
02 · Supabase / Neon free (auth + Postgres)$0.00
03 · Cloudflare R2 (cache + blobs)$1.00
04 · Domain$1.00
05 · LLM API calls (proxy passthrough)??? — scales with usage
06 · Monitoring (Sentry / basic observability)$0.00
07 · Edge function execution (Vercel/Cloudflare Workers over free tier)$25.00
TOTAL / mo$27.00 + usage
▸ break-even:immediately — if their plan is higher than $1/mo you can run a minimal rival for less; this is a cost-play at indie scale.
build

what you're up against.

48 hours to wire proxy + basic routing · 24 hours to add billing/spend tracking UI · 3 days polishing docs, SDKs and example apps · 3 days for testing, deploy and a Framer/landing page clone
easy
medium
hard
nightmare
01
easy
OpenAI-format request passthrough
Map incoming OpenAI-style payloads to provider-specific calls; straightforward mapping and small schema changes.
02
easy
Auth & API key management
Store keys, rotate, and allow per-team/provider keys; use Supabase auth or simple JWTs.
03
medium
Spend tracking and reporting
Collect usage and cost per model/provider; requires reliable metrics and clear UI for teams.
04
medium
Load-balancing & failover
Route to cheaper models, retry on failures, and surface degradations; needs thoughtful fallbacks.
05
hard
Latency and streaming support
Implementing streaming proxies and minimal buffering to preserve model streaming semantics is fiddly and easy to get wrong.
06
hard
Security and provider credential safety
Convincing teams to route keys through your service needs audited practices and clear isolation; not SOC2-level but still sensitive.
stack

their position.

detected signals· measured
cmsFramer
recommended stack · inferred
inferCloudflare Workers (or Vercel Edge Functions)inferSupabase (auth) + Postgres (usage logs)inferCloudflare R2 (cache)inferResend / Postmark free tier (email for alerts)inferFramer / simple landing page
rivals

who else has tried this.

option A
Open-source gateways (e.g. Tyk/Local proxy or OSS LLM gateway)
self-host a simple reverse-proxy that normalizes OpenAI-format requests; no vendor lock-in.
option B
Managed API bridges (Anthropic/Replicate adapters)
use provider-specific adapters or existing bridges when you need just one extra model or better cost control.
option C
Lower-tech substitute: environment-specific wrappers
a tiny library in your stack that switches between providers using feature flags and client-side keys — avoids a central gateway.
compare

similar scans.

same shape - different moat
ready to wedge in?
Get the wedge plan. Ship a contender by Sunday.
▸ generated with love, by a heartless robotverdict v2.5 · saaspocalypse.dev