Integrations

Aionis connects to agent hosts, file-reading agents, SDK clients, MCP clients, and external memory backends. Pick the integration path based on how much control you have over the loop.

For the shortest real Agent connection path, start with Connect an Agent.

Choose A Path

TypeScript SDK

Use it when you control a Node or TypeScript Agent loop and want the full observe -> guide -> feedback -> measure -> snapshot path.

Open TypeScript SDK

Raw HTTP

Use it when your host is not TypeScript or you want language-neutral calls.

Open Raw HTTP

AIFS file surface

Use it when your agent reads workspace files more reliably than it calls tools. AIFS mirrors governed guide context, active path, blocked memories, rehydrate pointers, receipts, and snapshots into .aionis/.

Open AIFS setup

Substrate

Use it when you want durable evidence, stronger ordinary-memory recall, backup, migration, preview context, or optional Zvec candidate search while keeping Runtime guide decisions governed by Aionis.

Open Substrate setup

MCP bridge

Use it when Cursor, Zcode, or another MCP client should talk to Aionis.

Open MCP setup

Claude Code plugin

Use it when you want Aionis context before prompts and execution evidence after tool use.

Open Claude Code setup

Multi-agent hosts

Use it when planner, worker, verifier, reviewer, or handoff roles share memory.

Open Multi-Agent Hosts

External memory

Use it when Mem0, Zep, Supermemory, vector DBs, markdown, logs, or internal memory already produce candidates and Aionis should govern admission.

Open External Memory

Core Contract

Every integration eventually maps onto the same product loop:


observe -> guide -> agent action -> feedback -> measure -> snapshot

Your host owns execution. Aionis owns memory admission, context compilation, feedback attribution, measurement, and audit replay.

Integration Decision Tree

Question	Recommendation
Do you control the Agent loop code?	Use the SDK.
Do you want Claude Code to use Aionis automatically?	Use the Claude Code plugin.
Should the Agent read Aionis through files in the workspace?	Use AIFS.
Do you want stronger factual recall and durable evidence search?	Add Substrate.
Does your Agent host expose MCP tools?	Use MCP.
Are you integrating from Python, Go, Rust, or a service boundary?	Use raw HTTP.
Do multiple Agents pass work across sessions?	Use the multi-agent host contract.
Do you already have a memory backend?	Keep it and add Memory Firewall through external candidates.

Data You Should Preserve

Whichever path you choose, keep these fields stable:

Field	Why
`tenant_id`	Logical tenant boundary.
`scope`	Project or workspace boundary.
`run_id`	Execution session identity.
`task_signature`	Stable workflow or task identity.
`agent_id`	Agent that produced or consumed evidence.
`team_id`	Shared lane for multi-agent memory.
`guide_trace_id`	Feedback attribution key.

Most integration failures come from identity and scope mismatches.

Execution Responsibility

Aionis provides memory admission, context compilation, feedback attribution, and audit replay. Your host provides shell access, browser access, CI access, model calls, and tool execution.

Start with Quickstart, then pick the path above.