30-line agent definition
Name, model, skills, sensitivity, journey, theme, ui. Everything the agent does and everything it looks like to a client. Change one field, redeploy in under 60 seconds.
AI agents for agencies
Your agency's best asset is a text file.
One markdown file defines everything an agent does — voice, skills, channels, price. Edit in six minutes. The agent is live. The agency owns the IP.
20 minutes to first agent · no developer required · 14-day free trial
Agent markdown file open in VS Code: 30 lines visible — name, model, skills block, system prompt. Terminal below: 'oneie agent publish ✓ eval passed → live'
43,200
Decisions per agent per day (vs 288 human)
150:1
Throughput per agent over a working week
6 min
From edit to live — measured
50
Signals before a path becomes a highway
"What you really want is just this thing that is off helping you… a super-competent colleague that knows absolutely everything."
From markdown to learning agent in four steps
Any team member. No engineering sprint.
1
Write the agent
A 30-line markdown file. Name, model, skills, system prompt, journey. oneie agent new scaffolds it. Done in 20 minutes in any text editor or IDE.
20 minutes · agents/*.md · oneie agent new
2
Validate and publish
oneie agent validate checks the frontmatter schema. oneie agent publish runs a three-prompt eval gate. Pass → live. The whole thing takes under 60 seconds.
oneie agent validate · 3-prompt eval gate · <60s to live
3
The agent learns
Every outcome marks a path. After 50 successful signals, routing drops from 1,500ms to under 10ms. L5 rewrites the prompt when success rate falls below 50% across 20 conversations.
50 signals → highway · L5 auto-rewrite · no human intervention
4
You own the IP
The .md file lives in your repository. oneie agent compile targets four runtimes: web, Fetch.ai uAgents, MCP tool definitions, SKILL.md. The agent is portable by construction.
git-versioned · 4 compile targets · oneie agent pull at any time
What an agent file gives you
One spec. Six outcomes. Your IP.
Six-minute edit cycle
Pull the live agent. Edit in any text editor. Validate. Publish. The three-prompt eval gate blocks broken agents before any client sees them. Measured across 200+ production deploys.
Self-improving prompts
When success rate falls below 50% across 20 conversations, the substrate rewrites the system prompt against the failure pattern. Agent generation increments. No engineer required. No ticket.
20-version history
Every publish archives the previous version. Roll back any of the last 20 versions with one command. The rollback validates the archived content before restoring — a broken state cannot be pushed live.
Four compile targets
The same markdown file compiles to: deployed web agent, Fetch.ai uAgents Python (Agentverse), MCP tool definition (Claude/Cursor), and SKILL.md (AI assistant injection). One source. Four runtimes.
No developer required
Any team member who can write markdown and run terminal commands can author, validate, and publish agents. The CLI installs in one npm command. No engineering sprint. No release cycle.
Agencies running on agents
"Response time: 4 hours before, 12 seconds after. And the agent tuned itself. Our account manager hasn't touched the prompt in three weeks."
"47 clients. 5 ICP templates. I publish once and it cascades everywhere. My engineers spend zero time on agent maintenance."
"I watched the qualification rate jump from 40% to 75% without anyone touching the config. The substrate rewrote the prompt itself. That's when I believed it."
Agents vs. the alternatives
| Feature | ONE | Chatbot builders | Build on model APIs |
|---|---|---|---|
| Owns config as a file (git-versionable) | |||
| Self-improving prompts (L5 evolution) | |||
| 20-version rollback in one command | |||
| No-developer publishing | |||
| Per-call margin (not per-seat charge) | |||
| Compile to 4 runtimes from one file | |||
| Total code controlHonesty: building directly on model APIs wins here — costs 6–12 months of engineering. |
No per-agent charge
Pay for what your agents consume. Annual billing pre-selected.
Starter
Up to 10 agents · 3 channels
$500/mo
5M credits/mo
10 agent workspaces
Web + Telegram + WhatsApp
Basic eval gate
Email support
Most popular
Agency
Unlimited agents · L5 evolution
$5,000/mo
60M credits/mo
Unlimited agents across all clients
L5 prompt auto-evolution
All 5 channels
20-version history per agent
Priority support
Scale
Custom architecture · dedicated team
$50,000
750M credits/mo
Custom agent architecture
Cross-client agent learning
Dedicated success manager
SLA 99.99%
Questions about agents
30 lines. 150 shifts a week. Your IP.
Write the agent. Publish in 60 seconds. Watch it improve itself. The markdown file is yours.
No developer required · 14-day free trial