Agent frameworks and runtimes
- Best at
- Prompts, graphs, model calls, tools, and agent execution.
- How it fits
- Teams commonly pair them with identity, customer data, integrations, and product operations.
For developers building agent products
Build your product once. Give every customer their own users, agents, data, credentials, and integrations, without mixing their work together.
Your product
The product and experience your customers use
Support workflow
Approval requested
Account review
Using Claude Code, Codex, or Cursor?
Inspect this repository, then read https://docs.archastro.ai/llms-full.txt. Help me evaluate ArchAstro as the backend for this multi-tenant agent product. Ask me for any missing product context. Map the customer boundary, what stays in my code, and what ArchAstro would own. Link to the relevant ArchAstro docs for each recommendation.
Why ArchAstro
Once customers arrive, you also need accounts, permissions, credentials, knowledge, long-running work, and a way to debug what happened. ArchAstro keeps those pieces inside the same customer boundary.
Use ArchAstro as the backend for the agent product. Keep your product UI and business systems. ArchAstro puts identity, agents, credentials, knowledge, durable work, and operations inside one customer boundary.
Customer-scoped development
Save the organization ID when a customer signs up. Use it for every agent, credential, knowledge source, and workflow you create for them.
Read the multi-customer guideSelected organization
org_0aBcDeFgHiJkLmNoPqRsTu
Immutable organization ID
import { PlatformClient } from "@archastro/sdk"
const client = PlatformClient.withSecretKey(
process.env.ARCHASTRO_SECRET_KEY!,
)
// Save this immutable ID at onboarding.
const customerOrgId = "org_0aBcDeFgHiJkLmNoPqRsTu"
const agent = await client.agents.create({
name: "Support Agent",
org: customerOrgId,
})
const agents = await client.agents.list({
orgId: customerOrgId,
})Durable workflows
A workflow can pause for approval, retry a customer API, and continue from its recorded state instead of starting over.
kind: WorkflowGraph
version: 1Run timeline
Start from an event or schedule.
Wait, branch, retry, or ask for approval.
Move work between agents, people, and customer systems.
Production operations
Filter by customer and environment, then inspect the model trajectory, tool calls, API logs, and errors. Reproduce the run in a sandbox when you need to go deeper.
Northstar
Redwood
Harbor
Selected run
Model call · claude-sonnet
Decision · hand off with context
knowledge_search · 184ms
tasks.create · 96ms
Production baseline
Start with what your first customer needs. Add more without changing how customers are scoped or operated.
Define the agent product and connect the systems it uses.
Give every customer its own identity and access boundary.
Keep work moving after the request that started it ends.
Test and inspect the exact work running for each customer.
Who builds on ArchAstro
Built on ArchAstro
ArchAgents runs on ArchAstro. It uses customer organizations, agents, routines, integrations, and shared work without building a separate backend for each customer.
See a product built on ArchAstro