Photo - ContextLane
64652

ContextLane

Self-hosted AI control plane: routing, policy, cost tracking

Cyprus
Market: Other, Artificial Intelligence
Stage of the project: Operating business

Date of last change: 07.11.2026
Go to the owner's profile
3
equalizer from 2000
help
Calculated and estimated occupancy of the project (more about ratings)
My rating
1
2
3
4
5
6
7
8
9
10
Average rating:
 

Idea

ContextLane helps companies scale AI without losing control of cost or quality. It routes each request to the lowest-cost capable model, enforces policy, and makes every routing decision auditable, so finance sees the saving and engineering sees why a request took a route. It is self-hosted: it runs in the customer's own infrastructure with their own provider keys.
In one fintech pilot it cut spend by about 74% against the customer's premium-model baseline.

Current Status

Current status. The product is built end to end: the self-hosted gateway, a customer admin UI, Chat, a developer CLI, a vendor portal and billing. It supports 8 model providers.

Traction. Two pilots are running on different workloads: a fintech company and an AI-agent development team. The fintech pilot is fully measured: over three months (June to August) it saved $20,069 against $7,112 of actual spend, about 74% against the customer's own premium-model baseline. That is roughly $6,700 saved per month against a $990 monthly fee.

Engineering proof. We used about 2 billion tokens to calibrate quality-aware routing across providers, and a synthetic gateway test at 50 requests per second ran for 5 minutes with 0% errors.

What is not true yet. No pilot has converted to a paid subscription. Converting both is our first milestone.

Market

Who we sell to first. Companies that already have a large and growing LLM bill and want control over it: AI-heavy software companies, fintech, legal tech, HR tech, and internal AI platform teams in larger companies. The buyer is usually the head of engineering or platform, with finance as a second stakeholder who wants to see the saving. We start with one high-volume workflow per customer, measured over a 3-month pilot with written success criteria.

Why they need us. These companies send most requests to a flagship model by default, cannot say which ones needed it, and often cannot send prompts or keys to a third-party service. Self-hosting and their own provider keys remove that objection.

Market size (our estimate). Enterprises spent about $12.5B on foundation-model APIs in 2025, part of $37B in enterprise generative AI spend (Menlo Ventures), and 20% of EU enterprises already use AI (55% of large and 30% of medium ones, Eurostat 2025). We divided total API spend by the yearly LLM spend of a company our size, using $109K (our fintech pilot) to $240K (conservative), and multiplied by our net revenue of about $9,190 per customer per year. That gives a TAM of $0.5–1.1B per year. We assume 40% of it sits in the US and EU companies we can reach first, a SAM of $0.19–0.42B.

Where we are going. Our base plan reaches about 59 customers and $0.57M ARR by month 36, roughly 0.1–0.3% of that SAM.

Problem or Opportunity

The problem. Companies send almost every AI request to their most expensive model. I saw this first-hand at a financial-services company where I work: the monthly LLM bill went from about $2,000 to $20,000 in eight weeks. We were not using AI ten times more. Summaries, drafts and one-line code questions all defaulted to a flagship model, although most of them could have run on a cheaper one. Public list prices of the providers we support differ by more than 100 times, so a wrong default is expensive.

Why it is getting worse. Enterprise spend on generative AI reached about $37B in 2025, 3.2× the year before (Menlo Ventures), and 51% of professional developers use AI tools every day (Stack Overflow 2025). Every new team and workflow adds more requests that follow the same expensive default.

The opportunity. Finance cannot see which requests needed a strong model, and engineering has no safe way to send simple ones elsewhere. We built a layer that picks the cheapest capable model for each request, under the company's own policy, and records every decision. That turns the waste into savings the company can measure.

Solution (product or service)

What we make. ContextLane is a self-hosted control plane for production AI. It sits between a company's applications and its model providers and decides, for every request, which model should answer it. It ships as one bundle: an OpenAI- and Anthropic-compatible API gateway, a team Chat, a developer CLI, and an admin UI for policy, savings and audit history.

How it helps a client.

Lower AI bills. Each request goes to the cheapest connected model that can do the job. In a fintech pilot this saved about 74% ($20,069 against $7,112 actual spend) compared with the customer's own premium-model baseline.
Visibility for finance. Every request is recorded with its real cost next to what the baseline model would have cost, so the saving is measured, not estimated.
Control for engineering and security. Companies set allowlists, exclusions and pinned routes, and can see why each request took its route. Routing and failover across 8 providers are handled by the gateway.
No new data risk. It runs in the customer's own infrastructure with their own provider keys. Prompts, keys and logs never leave their network, and our servers are not in the request path.
Fast to adopt. Teams change one base URL. No change to prompts or response handling.
How it is sold. One flat price of €990 per month (VAT included) with unlimited requests, seats and workspaces, after a free 30-day pilot with a savings report on the customer's real traffic.

Competitors

Most companies today do one of four things:

Do nothing. They default every request to one flagship model. This is our real competitor, and it is how a monthly bill can go fr om $2,000 to $20,000 in eight weeks.
Model gateways and marketplaces (for example OpenRouter, LiteLLM, Portkey). They give one API to many models and handle access, keys and fallbacks. Some are hosted by the vendor, some are open source.
Observability and evaluation tools (for example Helicone, Langfuse, LangSmith). They show what happened and how good the answers were, but do not decide wh ere a request should go.
Build in-house. A team writes its own proxy and rules. It works at first, but needs ongoing engineering and rarely keeps routing quality or the savings records up to date.
How we are different. Others solve model access, gateway plumbing or observability. We focus on what finance asks for: route each request to the cheapest capable model, enforce the company's policy, keep session context when the model changes, and prove the saving per request. We also run inside the customer's own infrastructure with their own keys, which matters to regulated companies that cannot send prompts to a third-party service.

What protects us. Our routing is calibrated on about 2 billion tokens across providers, and every decision is stored next to what the baseline model would have cost. We do not claim that gateways cannot add routing. Our bet is that a cost-control workflow built around measured savings and self-hosting is hard to copy as a side feature.

Advantages or differentiators

Savings we can prove. Every request is recorded with its real cost next to what the customer's baseline model would have cost, together with the routing decision behind it. In our fintech pilot this showed about 74% savings ($20,069 against $7,112). Finance sees the number, engineering sees why.
Self-hosted by design. The gateway runs in the customer's own infrastructure with their own provider keys. Prompts, keys and logs never leave their network, and our servers are not in the request path. We only receive licence checks and aggregate health totals.
Routing built around a quality bar. Each request is classified and sent to the cheapest connected model that clears the bar: routine tasks to cheap models, hard ones to strong models. Our routing is calibrated on about 2 billion tokens across providers, with tier-aware and cached-input pricing.
Policy control. Companies set allowlists, exclusions and pinned routes, and keep an audit trail of every decision.
Session continuity. Session context is preserved when the model changes, so a cheaper model can take over mid-conversation without breaking it.
Fast to adopt, simple to buy. Teams change one base URL, no change to prompts or response handling. It works with 8 providers, and one flat price covers unlimited requests, seats and workspaces.
What makes it hard to copy. Routing quality improves with measured data from real customer traffic, and the savings records and audit history become part of the customer's finance and security workflow. Self-hosting also means we earn trust that a hosted service cannot easily match in regulated companies.

Finance

Revenue streams.

Subscription (today). A flat €990 per month per customer, VAT included, billed monthly with no long-term contract. It includes unlimited requests, seats and workspaces. Customers keep paying their model providers directly with their own keys, so we do not resell tokens or add a margin on them.
Share of measured savings (planned, from month 13). An optional fee of 10% of the savings ContextLane measures for the customer. We plan for 15% of customers to adopt it, with a 50% haircut on the savings seen in our fintech pilot (about $6,700 per month).
After 19% VAT and about 6.7% payment fees, we keep roughly 77% of the price. Our base plan reaches about €766 net per customer per month. A 30-day free pilot comes first, so revenue starts only when a customer converts.

Cost structure today (about $15,750 per month). Routing-quality work $10,000 (evaluation of models and routing decisions, the largest line, which we expect to fall as quality improves), salaries $2,000 for the whole team of six, sales $3,000, infrastructure $500, legal and accounting $250. The founder takes no salary.

Cost structure after the round. About $25,000 per month goes to marketing and sales, mostly hires and outsourced outreach. Over the first 12 months we plan to spend about $430K: sales and marketing 70%, routing quality 23%, salaries 6%, infrastructure and legal 2%.

Margins. The product is self-hosted, so our hosting cost per customer is close to zero. Our main variable costs are payment fees and the cost of winning a customer: about $8.9K per paying customer in the base plan against about $25.5K lifetime revenue (plan, not yet observed).

Business model

Business model. B2B software sold as a self-hosted subscription: €990 per month per customer, unlimited requests, seats and workspaces, monthly billing. A customer starts with a free pilot on one real workflow, receives a savings report on its own traffic, and converts if the saving is clear. We plan to add an optional fee of 10% of measured savings from month 13. The customer keeps its own provider contracts and keys, so we do not resell tokens.

Channels.
Outbound sales (main channel). Marketing and sales hires plus outsourced outreach target AI-heavy software companies, fintech, legal tech, HR tech and internal AI platform teams. Our base plan assumes about 1,400 prospects contacted per month, 2% of them become meetings (28), 25% of meetings become pilots (7), and 40% of pilots become paying customers (about 2.8 per month). Budget is about $25K per month after the round.
Our website (supporting channel). A public savings calculator and guides let a prospect estimate savings with their own providers and token volumes before talking to us.
Land with one workflow. We start with one high-volume workflow and one team, prove the saving over the pilot, and expand from there.
Metrics we track and plan against.

Measured today: 2 active pilots; in the fintech pilot $20,069 saved against $7,112 actual spend (about 74%); 2B tokens used to calibrate routing; 50 RPS gateway test with 0% errors.
Planned (not yet observed): pilot-to-paid conversion 40%, monthly churn 3%, CAC about $8.9K, LTV about $25.5K (LTV/CAC 2.9×), payback about 11.7 months.
ARR plan, net of VAT and payment fees: about $194K at month 12, $417K at month 24 and $565K at month 36. With the round we expect 17 paying customers and about $152K ARR by month 10.
First milestone: convert both current pilots to paid subscriptions.

Money will be spent on

Use of funds. We are raising $300K to turn our pilots into paying customers. Our plan for the first 12 months is about $430K of spending, covered by the round and early revenue.

Sales and marketing, about 70% (roughly $300K, or $25K per month). A marketing hire and a sales hire, outsourced outreach and tooling. The goal is about 7 new pilots per month.

Routing quality, about 23% (roughly $97K). Evaluation of models and routing decisions, so routing stays accurate as providers and models change.

Salaries, about 6% (roughly $24K). The current team of six at $2,000 per month in total. The founder takes no salary.

Infrastructure and legal, about 2% (roughly $9K). Infrastructure of about $500 per month, plus legal and accounting of about $250 per month.

What this buys. About 10 months of runway in our base plan, including revenue, and about 17 paying customers and $152K ARR by month 10. A full 12 months needs about $336K, so we plan to start the next round before the runway ends.

Offer for investor

What we offer. We are raising at least $300K pre-seed through a SAFE (or a convertible note, if an investor prefers). We are happy to discuss the valuation cap and other terms directly with each investor.

What the money delivers. With $300K we plan to convert our two pilots into paying customers and reach about 17 paying customers and about $152K ARR within about 10 months, with sales and marketing taking most of the budget. A larger round extends this runway toward 12 months and beyond. This should position us for a seed round on real revenue and measured conversion data.

What investors get besides the SAFE.

Pro rata rights. The right to invest in our next round to keep their share.
Regular reports. A regular update on pilots, paying customers, cash and the next milestones, so investors see progress as it happens.

Team or Management

Risks

1. No customer has paid yet. Two pilots are running and one is fully measured, but none has converted to a paid subscription. Our plan assumes 40% of pilots convert, based on industry ranges, not our own data. What we do: pilots run on one real workflow with written success criteria, and converting both is our first milestone.

2. Sales assumptions are unproven. Our funnel (2% of outreach becomes a meeting, 25% of meetings become pilots) is an estimate. What we do: we start with narrow segments, track each step from the first month, and upd ate the plan from real conversion data.

3. Falling model prices can shrink savings. If flagship models get much cheaper, the gap between a flagship and a cheap model narrows. What we do: our value is not only price arbitrage but also policy control, audit history and measured savings per request, and these stay useful at any price level.

4. Routing quality. Sending a request to a cheaper model can lower answer quality. What we do: routing is built around a quality bar, customers can pin routes and se t allowlists and exclusions for sensitive workloads, and savings are measured against the customer's own baseline. We invest the largest share of our current budget (about $10K per month) in routing quality.

5. Competitors may add routing. Gateways and observability tools can build similar features. What we do: we focus on a cost-control workflow with measured savings and self-hosting, and on routing calibrated on real traffic (about 2B tokens so far).

6. Small team and short runway. We are a team of six plus one founder, with salaries well below market and the founder not paid. Current savings cover about 4 months at today's burn, and a $300K round covers about 10 months in our base plan. What we do: the round is sized to reach paying customers, and we plan to raise the next round before cash runs low.

7. Deployment and support effort. A self-hosted product has to be installed and supported inside each customer's environment. What we do: the pilot includes help with installation, and the gateway runs on the customer's own Postgres and Redis.

8. Pricing. A flat price captures only a part of the value we create. What we do: we plan an optional share of measured savings from month 13, once we have seen real results.

Incubation/Acceleration programs accomplishment

We didn't participate in other programs yet
5,00
1
2
3
4
5
2 voices
Sign in/Sign up
arrow_back
EN
more_horiz
close
visibility170
star0
Add to favorites
Delete from favorites
share
close
thumb_up0
Like
Unlike
Idea
Current Status
Market
Problem or Opportunity
Solution (product or service)
Competitors
Advantages or differentiators
Finance
Invested in previous rounds, $
Business model
Money will be spent on
Offer for investor
Team or Management
Mentors & Advisors
Lead investor
Risks
Incubation/Acceleration programs accomplishment
Won the competition and other awards
Invention/Patent
Photos
Product Video
Presentation