Plan memory federation with the Praxa CLI

The 0.3.0 memory commands are local configuration tools. They do not connect to a provider, read memory, write memory, synchronize records, import data, mirror data, or execute a cutover.

Supported providers

CLI value	Backend client you supply later
`mem0`	Mem0 client
`zep`	Zep client
`graphiti`	Graphiti transport
`langgraph`	LangGraph store
`letta`	Letta blocks/messages client
`openai_agents`	OpenAI Agents session resolver

The SDK adapters remain credential-blind. Your backend creates and owns the provider clients.

Preview a source

praxa memory source add mem0 \
  --mode federated \
  --project-dir . \
  --dry-run \
  --json

The response identifies the provider, a generated source ID, read_only access, and local_configuration_only execution. Because this is a dry run, the project remains unchanged.

Write local source metadata

After reviewing the dry run:

praxa memory source add mem0 \
  --mode federated \
  --project-dir . \
  --json

Review the resulting .praxa files. They must not contain an endpoint, provider credential, secret reference value, tenant token, or memory content.

Inspect the sync plan

praxa memory sync plan \
  --project-dir . \
  --dry-run \
  --json

The plan is intentionally non-executable. It describes configured sources and reports executable: false.

There is no memory sync run, memory mirror, or memory cutover operation in 0.3.0. The CLI refuses those commands instead of implying data moved.

Connect the source in code

The CLI metadata is not a provider client. Implement the corresponding server-side adapter explicitly:

import {
  MemoryFederation,
  createMem0MemorySource,
} from "@praxa/sdk/memory";

const source = createMem0MemorySource({
  id: "mem0-primary",
  client: backendOwnedMem0Client,
  mapNamespace: ({ tenantId, subjectId }) => ({
    userId: `${tenantId}:${subjectId}`,
  }),
});

const federation = new MemoryFederation({ sources: [source] });

Use explicit tenant and subject namespace resolution. Never accept a provider namespace directly from an untrusted frontend.

Verify the boundary

Run source add --dry-run in an empty temporary directory.
Require JSON output and zero files written.
Apply the source and inspect every generated file.
Search for credential-shaped strings; require zero matches.
Run sync plan --dry-run and require executable: false.
Attempt memory sync run, mirror, and cutover; require refusal.
Unit-test the SDK adapter with a fake provider client.
Run a disposable provider canary and prove tenant/subject isolation, provenance, partial degradation, and no provider writes.

Troubleshooting

Symptom	Meaning	Fix
`--mode federated is required`	Mirror/cutover is unsupported	Use the read-only federated mode
Provider rejected	Value is outside the six exact identifiers	Use the CLI spelling from the table
Plan contacts no provider	Expected	Build and test the SDK adapter separately
`executable: false`	Expected safety boundary	There is no sync execution in this release
Source returns no recall later	Provider client or namespace resolver issue	Test provider access and mapped tenant/subject values in the backend

Memory federation SDK

Build the backend adapter and inspect portable kinds, provenance, limits, and failure states.

End-to-end memory tutorial

Configure real provider clients and test partial degradation and isolation.

Workflows

Automate and operate

Plan memory federation with the Praxa CLI

Supported providers

Preview a source

Write local source metadata

Inspect the sync plan

Connect the source in code

Verify the boundary

Troubleshooting

Memory federation SDK

End-to-end memory tutorial

​Supported providers

​Preview a source

​Write local source metadata

​Inspect the sync plan

​Connect the source in code

​Verify the boundary

​Troubleshooting

Memory federation SDK

End-to-end memory tutorial

Supported providers

Preview a source

Write local source metadata

Inspect the sync plan

Connect the source in code

Verify the boundary

Troubleshooting