Agent workspace
RYL
The agentic development harness. Tasks, runs, tool calls and approvals.
I am Legion. We are Legion.
We build software solutions for teams who work with AI agents.
legion stack
OSPRY Live Find out who is on your site WHIPT Private beta Run your company with AI agents rflectr Windows beta Record what your coding agents do knectar Sign-ups soon Keep your AI keys in one placealsoRYL, the apiary, shrke, legioncode.cloud
What we make
Legion Code is a software company in Cleveland, Ohio. We build for teams who work with AI agents: find the buyers on your website, run your company with agents, see what your coding agents do, and keep your AI keys safe.
Website visitors Live
I am OSPRY.
I find out who is on your website.
Website visitors
Real buyers read your pricing page and leave no name.
OSPRY names them, scores how serious they are, writes the brief and puts them in your CRM while they are still warm. Your rep calls while they are still reading.
visit/pricing · 3rd visit · 41s
namedDana Chen · Archer Labs
emaild.chen@
score86 / 100
briefread pricing twice, opened the HubSpot page
crmsent · HubSpot contact + deal
How OSPRY works
- Snippet Paste one line of code. Every page and every visit, first-party.
- Score Each visitor scored 0 to 100 against your best kind of customer.
- Brief Name, work email, LinkedIn and company, plus what to open with.
- CRM A CRM row and a Slack ping, minutes after the visit, not Monday.
AI teams Private beta
I am WHIPT.
I run your company with AI agents.
AI teams
Your competitors have a folder of prompts. You have a company.
Custom GPTs cannot be given a budget. Freelancers cannot be audited. WHIPT hires AI agents by what they can do, hands them real work, stops what should not leave the building, and shows you the bill for every run.
taskdraft Q4 renewal emails
agentComms-2 · hired with a budget
drafted38 emails in 6 seconds
gateoutbound email · waiting for you
sent0 until you say so
receipttool calls, tokens, cost, approvals
How WHIPT works
- Hire Pick a capability, a model and a budget.
- Assign Hand it a task like you would a teammate.
- Gate Outbound email waits for you. 38 emails drafted in 6 seconds, 0 sent until you say so.
- Audit Every run is a receipt: tool calls, tokens, cost and who approved what.
Coding agents Windows beta
I am rflectr.
I record every request your coding agents make.
Coding agents
Your coding agents make requests you never see.
rflectr is a desktop app that keeps a record of every request your coding agents make, and lets you send any of them to a different provider. Your login stays yours.
Claude Code Observe 312 ms · 6k tokens · no prompt kept
Codex CLI Observe 488 ms · 4k tokens · no prompt kept
Cursor Off not in the path
Codex CLI Route to your provider · 290 ms · 3k tokens
every agent login untouched
How rflectr works
- Install it Windows 11 now, macOS 14 soon. Every agent starts Off, so rflectr is not in its path.
- Turn one on Pick an agent and set it to Off, Observe or Route. Observe watches it. Route sends it where you say.
- Read the record One row per request: model, latency, tokens, cost. Never the prompt. Routed requests go to the provider you set, on your keys.
AI keys Sign-ups soon
I am knectar.
I keep all your AI keys in one safe place.
AI keys
Your OpenAI key sits in every app.
One leak means pasting a new key into all of them. With knectar, each app holds only its own knectar key. Switch off the one that leaked and nothing else breaks.
.envOPENAI_API_KEY=your knectar key
requestopenai/gpt-4o-mini
openaitimed out
backupanthropic/claude-haiku-4-5 · 200
logone row · model, time, tokens, cost
your keynever left knectar
How knectar works
- Swap the key Your app holds only its knectar key, in the same OPENAI_API_KEY spot.
- knectar sends it knectar checks the knectar key and adds your OpenAI key. Your app never sees it.
- Backup If OpenAI fails or times out, the same request goes to the backup model you picked.
- Logs One row for every AI request, with the model, time, tokens and cost.
02 The family
Twelve products. One language.
Each on its own domain. Each speaks Legion. OSPRY leads.
Identity resolution, by Legion.
One snippet. Which companies, and which people, visited your site, what they did, and an AI sales brief on each, pushed to your CRM. Observe. Score. Profile. Reach. Yield.
RYL
The agentic development harness. Tasks, runs, tool calls, approvals. One lion that changes state as the run does.
rflectr
A record of every request your coding agents make. Per harness: Off, Observe or Route. Your login stays yours.
shrke
The control plane as a web app plus tray client. Provider, model, latency, tokens and cost on every request.
knectar
The AI gateway. One endpoint, every provider, fallback chains, virtual keys, cost per request.
WHIPT
The company employee AI. Browser, comms, code, research and media, working the business while the harness works the code.
legioncode.cloud
Where every Legion product runs. Cloudflare at the edge, sandboxes for the agents, one tailnet for the humans.
MACSITNA
Anti-scam intelligence network. Anti-scam spelled backwards. Lawful. Transparent. Evidence-driven.
Owl's Roost
Indigo ground, gold for wins. Live now.
OnlyFacts
A research library where every claim carries its evidence class. Ask Susan.
Time Debt
Four minutes of questions, then the invoice: hours lost a week and a Time Debt Score from 300 to 850.
the apiary
One memory. A whole hive around it. Shared, lasting memory for your coding agents, on hardware you control. With Activeloop.
We are Legion.
Four products from one small team, made for teams who work with AI agents.
The layers underneath.
Memory
the apiary
Shared memory. Learn something once and every agent recalls it.
Control panel
shrke
The control plane as a web app and a tray client.
Hosting
legioncode.cloud
Where every Legion product runs.