Infrastructure
for production agents
A runtime and control plane for AI agents. Persistent identity, long-term memory, computer use, integrations, workflows, and multi-tenant isolation. Deploy from Claude Code, Codex, or the CLI.
- One platform for the full agent lifecycle. Runtime, auth, integrations, observability, and workflows. Everything agents need to run in production.
- Persistent across sessions. Agents retain memory, follow scheduled routines, and maintain identity over time. Each organization’s data stays isolated.
- Cross-company coordination. Agent Network connects agents across organization boundaries through shared teams and threads, with each company’s data kept separate.
- Multi-tenant by default. Organization isolation, per-org credentials, sandboxed environments, SSO, and audit trails.
Two ways to build on ArchAstro
Full products with agents inside
End-to-end applications where agents handle core workflows: support products, financial tools, internal operations. Agents, memory, OAuth, webhooks, workflows, and a developer portal.
Build AgentsStandalone agents that coordinate across companies
Standalone agents with identity, memory, tools, and computer use. Deploy to Slack, Claude, Codex, or any surface. No app needed. Use Agent Network for cross-org collaboration with other companies’ agents.
Fastest way to get started
Just tell your coding agent.
Paste this into Claude Code, Codex, or Gemini CLI. It reads the docs and sets up everything for your stack.
Set up ArchAstro in this repo so we can deploy an agent and test it.
1) Read:
https://docs.archastro.ai/llms-full.txt
2) Ask me for any missing ArchAstro credentials or environment variables.
3) Install the ArchAstro CLI and run: archastro auth login && archastro init
4) Write an agent.yaml template (kind: AgentTemplate) with:
- a clear identity/instructions
- the participate preset routine (so it responds in conversations)
- search and knowledge_search builtin tools
- memory/long-term installation
5) Deploy it: archastro deploy agent agent.yaml --name "Support Agent"
6) Test it:
- create a thread, user, and send a test message
- OR create an agent session and exec a test prompt
7) When complete, summarize what was created and how to test it again.Prefer CLI? brew install archastro, then run archastro init.
What you’d otherwise spend months building.
Compared to model providers and consumer-first framework stacks.
What your agents get
Swipe to compare →
| Typical frameworks | DIY | ArchAstro | |
|---|---|---|---|
| Agent identity | In-process state or checkpoints | Custom state model | Persistent agents with routines, installations, and config versioning |
| Memory | Checkpoint-based or add-on vector DB | Build retrieval pipeline | Built-in long-term memory with vector search and auto-capture |
| Computer use | Not typically included | Provision infra yourself | Managed VMs attached to agents, operated from the CLI |
| Scheduled reasoning | Cron libraries or external schedulers | Multiple services | Cron routines with full LLM tool access, all with run history |
| Knowledge and search | Add-on retrieval pipelines | Multiple services | Hybrid search (semantic and keyword) with managed ingestion |
What your team gets
Swipe to compare →
| Typical frameworks | DIY | ArchAstro | |
|---|---|---|---|
| Multi-tenancy | Not included | Build it yourself | Org isolation, per-tenant RBAC, SSO, sandboxes, audit trails |
| Extensibility | Large plugin ecosystems | Custom adapters | MCP servers, webhooks, scripts, and workflows on one platform |
| OAuth and credentials | Manual per-integration | Custom glue code | Declarative OAuth with refresh and PKCE. Per-org credential isolation. |
| Operator tooling | CLI wrappers or dashboards you build | Build internal tooling | CLI and portal for agents, orgs, configs, sandboxes, impersonation |
| Coding agent DX | Not supported | Manual setup | Deploy from Claude Code or Codex via llms-full.txt in minutes |
| Impersonation | Not supported | Custom dev tooling | Adopt any agent's context locally, inspect tools, install skills |
What helps companies collaborate
Swipe to compare →
| Typical frameworks | DIY | ArchAstro | |
|---|---|---|---|
| Cross-org coordination | Not supported | Net-new architecture | Shared teams and threads across orgs, each keeping isolated data |
| Multi-agent collaboration | Multi-agent graphs within one process | Build coordination layer | Multi-agent teams with shared threads across orgs, not just one runtime |
| Realtime threads | Not typically included | Multiple services | Persistent threads with realtime messaging and event-driven agent participation |
What you get
Agent runtime + control plane. Any LLM.
Agent Identity and Routines
Persistent agents with managed lifecycle, event-driven routines, and per-tenant credentials. Schedule agents to reason and act on a cron with full access to their tools and knowledge.
Agent Network
Agents collaborate through teams and shared threads. Build multi-agent systems within your org or coordinate across organization boundaries.
Agent Memory
Long-term, vector-searchable memory that persists across sessions. Agents remember preferences, facts, and instructions automatically.
Computer Use
Provision remote machines and execute shell commands. Agents can check out repos, run builds, and interact with real infrastructure.
Knowledge and Search
Combine semantic and keyword retrieval so agents can reason over both intent and exact matches across all connected data.
Workflows and Scripts
AgentScript: an expression-oriented language with HTTP calls, JWT signing, email, Slack, and data pipelines. Visual workflows with 11 node types. Runs on the platform.
OAuth and Integrations
Declarative OAuth with refresh, PKCE, and device flow. GitHub App webhooks and installation-derived tools. Credentials are per-org and encrypted at rest.
Observability and Audit
Inspect routine runs, workflow execution, and agent activity. View run history, error details, and execution timelines from the CLI or portal.
Multi-Channel Delivery
Chat threads, email, scheduled jobs, apps, and CLI. Same agent, same identity, same tools everywhere.
Extensibility
Remote MCP servers, webhooks, scripts, and workflows to integrate any system. Each integration is org-isolated with its own credentials and access boundaries.
Cross-Org Coordination
Your agents and a partner’s agents coordinate through shared teams and threads. Each org keeps its own data. In pilot with select enterprises.
GitHub App Integration
Install GitHub Apps and auto-generate agent tools from integrations. Webhook events flow directly into agent routines.
Impersonation
Adopt any agent's context from your terminal. Inspect its tools, run them locally, and install its skills into Claude Code or Codex.
MCP Servers
Connect agents to remote MCP-compatible tool servers. Stripe, Notion, Sentry, Linear, and the broader MCP ecosystem.
Sandboxes and Evals
Isolated environments with captured emails and separate credentials. Eval framework for measuring agent quality. Test in sandbox, promote to production.
See it in action
Agents that remember, collaborate, and compute.
User
I need help with invoice INV-2041
Triage Agent
Billing request detected. Routing to specialist.
Billing Agent
Found INV-2041. Amount: $2,400. Status: overdue. I can generate a payment link or adjust the due date.
Multiple agents collaborate in shared threads via participate routines.
Prefers concise responses, uses metric units
Account tier: enterprise, region: EU-West
Always escalate billing disputes over $5k
Long-term memory persists across sessions with vector search and auto-capture.
$ git clone repo.git
Cloning into 'repo'...
done.
$ npm install && npm test
added 847 packages
42 passing (3.2s)
$ npm run build
Build complete.
Provision machines and execute commands through the CLI.
Reference implementation
See what a production app built on ArchAstro looks like.
Helper: a live product built on ArchAstro.
Helper is a production assistant used daily by happy users, built entirely on the ArchAstro platform. It demonstrates what’s possible when you combine persistent agents, team networks, and integrations.
Agent Network brings this model to cross-organization coordination, now in production pilot with enterprise companies.
usehelper.com →9:41
Inbox
Your daily briefing
Figma trial expires tomorrow. 2 meeting...
Permission slip submitted
Done! I filled it out and sent it back to s...
Flight rebooked
Moved you to the 3pm direct flight inst...
Agent Network
In pilot with select enterprisesYour agents and theirs, on trusted ground.
When agents need to work across company boundaries, neither side should have to trust the other’s infrastructure. Each company keeps its own org with isolated credentials, data, and execution environments. ArchAstro manages the shared layer: teams, threads, and context that both sides can see.
This enables workflows no single-org platform can support: cross-company agent coordination, shared knowledge across org boundaries, and human oversight at every crossing point. Multiple agents, multiple humans, multiple organizations, each keeping their own data private.
Company A
Trusted Layer
ArchAstro
Trust · Routing · Audit
Company B
Each company owns their org. ArchAstro mediates. Neither side exposes internal infra.
# Company A: create a shared team and invite a partner
archastro create team -n "Partner Coordination"
archastro create threadmember --thread <team_id> --agent-id <agent_id>
# Share the invite — Company B joins with their own agents
archagent auth login
archagent create threadmember --thread <shared_thread_id> --agent-id <agent_id>Developer Portal
Define behavior. Test it. Ship it.
Workflows connect triggers, AI agents, and actions into multi-step pipelines. Scripts handle the deterministic logic in between: routing, transformations, and branching. Both run and test inside the portal.
Trigger
thread.message_added
Script
Classify intent
Command
Draft reply
Command
Create task
HTTP
Notify Slack
Drag-and-drop. Test with live data.
kind: AIWorker
id: support_agent
version: 3
name: Support Assistant
analysis:
orchestrator:
kind: SingleAgent
agent:
model: openrouter/google/gemini-2.0-flash
response:
orchestrator:
kind: SingleAgent
agent:
model: anthropic/claude-sonnet-4-5-20250929YAML configs with validation, versioning, and schema checks.
// Workflow script: route messages to the right handler
let arr = import("array")
let str = import("string")
let subject = str.lowercase($.message.subject)
let labels = $.message.labels
let is_urgent = contains(subject, "urgent")
|| arr.some(labels, fn(l) { l == "priority" })
let route = if (is_urgent) {
"escalate"
} else {
if (contains(subject, "invoice")) { "billing" } else { "general" }
}
{ route: route, urgent: is_urgent }Output
{
"route": "escalate",
"urgent": true
}Valid
Built-in namespaces: array, map, string, math, result
AgentScript: JSONPath data access, closures, and inline testing.
Shipping production agents?
We love to spend time with other builders. Join our waitlist to stay connected, or drop us a mail at hi@archastro.ai if you just want to brainstorm and chat.
The agent platform
for serious production use.
Persistent identity, long-term memory, computer use, multi-tenant isolation, and cross-org coordination. We handle the infrastructure. You build the agents.