Open source
Context engineering
Context Agent
Build, route, validate, and evaluate the smallest project context an AI system needs for the decision at hand.
Open source
Five focused agents
Specialized AI agents for real product work
AgentsOS gives product teams clear owners for context, design, growth, safety, and communication work. Use one agent independently or combine them around the job you need to finish.
MIT licensed · bounded responsibilities · verifiable outputs
Route the smallest useful context
$ npx context-os route \
"How do we deploy to production?"
✓ route ready · 12 files · 8.4k tokensAudit a page without API keys
$ cd agents/marketing-agent
$ python3 scripts/audit_page.py \
https://example.com --format markdown
✓ readiness 100 / 100 · 13 checks passedConnect Mail Agent to Codex
$ codex mcp add mailagent -- \
npx -y -p @mailagent/mcp@0.2.16 mailagent-mcp
$ codex mcp list
✓ mailagent · connected
ContextInterfaceGrowthSafetyCommunication
Agent catalog
One clear owner for every kind of product work
Each agent has a bounded responsibility, its own implementation and tests, and an output you can inspect. Start with the bottleneck in front of you.
Design skill
Interface quality
Design Agent
Turn briefs, screenshots, and implementation constraints into interface specifications, critiques, and prioritized repairs.
New
Growth and discoverability
Marketing & SEO Agent
Turn product evidence into positioning, technical SEO audits, content systems, conversion experiments, and measurement plans.
CLI + reports
Software safety
Safety Agent
Audit cryptographic posture, post-quantum migration readiness, policy gates, and implementation risks with traceable reports.
Hosted + open source
Email and QA
Mail Agent
Create isolated inboxes, extract OTPs and magic links, automate signup tests, and give agents policy-gated workspace communication through MCP and REST.
How AgentsOS works
Specialize the work. Keep control of the result.
AgentsOS is organized around responsibility boundaries, not a promise that one model can do everything.
Start with the job
Select the agent that owns your current bottleneck. You do not need to install or understand the whole repository.
Use a bounded workflow
Give the agent the relevant evidence and constraints. Each agent keeps its own tools, references, and verification contract.
Review concrete output
Inspect the report, specification, audit, CLI result, or production action before combining it with the next agent.
The operating principle
The smallest capable agent
for the job
More tools and more context do not automatically create better work. AgentsOS keeps responsibility explicit so outputs stay easier to understand, test, replace, and improve.
Clear scopeRelevant evidenceVerification
Built to be inspected
Evidence before platform promises
The repository exposes what exists today, how it is tested, and where each product is still evolving.
Measured context routing
Cross-project evaluations cover 149 questions with documented token-efficiency results.
Deterministic audits
HTML and URL readiness audits run locally, include tests, and require no API keys.
Traceable safety reports
Scans, threat models, inventories, and policy checks become reviewable CLI and HTML reports.
Production infrastructure
Mail Agent runs on Cloudflare with a post-deploy gate while remaining self-hostable.
Available as a hosted service
Mail Agent is the first live module of AgentsOS Cloud
Add temporary inboxes and structured email actions to CI or an autonomous agent without scraping an inbox UI. Start hosted, or self-host the same MIT-licensed code.
// Cursor or Codex via MCP
await mailagent_verify_signup({
service: "github",
runId: "release-4821",
deleteAfterSuccess: true
});
// → OTP or primary magic link
// → structured confidence and evidenceFAQ
Before you start
AgentsOS is growing in public. The boundaries are intentional; the maturity of each agent is visible.
Is AgentsOS one general-purpose AI assistant?
No. It is a collection of focused agents with explicit responsibilities, inputs, outputs, and verification paths.
Do I need to install every agent?
No. Each agent can be used independently and keeps its own dependencies, tests, and release lifecycle.
Can the agents work together?
Yes. Combine their outputs in sequence—for example, route project context, review an interface, prepare positioning, then run a safety gate. You stay in control of the handoffs.
Which product is hosted today?
Mail Agent is the first hosted AgentsOS product. The other agents currently run from the open-source repository.
Is AgentsOS open source?
Yes. The repository uses the MIT License. Individual agents expose their source, documentation, and focused verification commands.
Where should I start?
Choose the agent that matches your immediate bottleneck. If you need production email verification or signup QA, start with Mail Agent.
Start with one clear job
Choose your first agent
Explore the source, run the focused verification, and combine agents only when the work calls for it.