Skip to Content
Aionis v0.3.2 is the current Runtime baseline for SDK/API hosts, MCP clients, plugins, Substrate-backed recall, and self-managed agent loops.
API ReferenceSDK & API Surface

SDK and API Surface Matrix

This page is the authoritative map of Aionis public surfaces. Use it to decide which API, SDK method, or integration surface to call, and which outputs are safe to pass to an Agent.

Status Labels

StatusMeaning
PrimaryRecommended product path for new integrations.
StableSupported public surface, but lower-level than the primary path.
IntegrationA package or plugin surface that delivers the same Runtime contract to a host.
Operator-onlyFor review, audit, debugging, or lifecycle control. Do not pass directly to an Agent.
AdvancedUseful for specific deployments or workflows. Not required for the first integration.
InternalRuntime support surface. Do not expose as an Agent-facing product API.
Do not buildExplicit non-goal. Do not create this as a product surface.

Recommended Entry Points

GoalUseAgent-facing outputStatus
Node or TypeScript Agent host@aionis/sdk guideAgentContext()result.agent_promptPrimary
Multi-agent or role-aware host@aionis/sdk execution.guideAgentContextForRole()result.agent_promptPrimary
HTTP-only hostPOST /v1/guideagent_context.prompt_textStable
MCP-capable host@aionis/mcp aionis_contextagent_promptIntegration
File-reading Agent@aionis/aifs refresh.aionis/guide.mdIntegration
Claude Codeaionis-claude-code hooksinjected Aionis contextIntegration
External memory backendPOST /v1/memory/govern or SDK governMemory()agent_context.prompt_textStable

If you are writing a new Node or TypeScript integration, start with:

const context = await aionis.guideAgentContext({ query_text: "Continue from the current accepted route.", context_mode: "compact_agent", }); await agent.run(context.agent_prompt);

Core Runtime API

SurfaceRuntime APISDK methodAgent-facingStatusNotes
Observe memory and execution evidencePOST /v1/observeobserve(), remember(), execution.observeStep()NoPrimaryRecords ordinary facts, execution steps, validation output, and handoffs.
Compile raw guidePOST /v1/guideguide(), execution.guideForRole()Yes, via agent_context.prompt_textStableHTTP-level governed context. Use mode: "full_power" and context_mode: "compact_agent" for compact execution-memory prompts.
Compile SDK Agent contextPOST /v1/guide plus resolve callsguideAgentContext()Yes, via agent_promptPrimaryAdds SDK execution contract and resolved inspect/rehydrate evidence around the guide.
Compile role-aware SDK Agent contextPOST /v1/guide plus resolve callsexecution.guideAgentContextForRole()Yes, via agent_promptPrimaryRecommended for planner, worker, verifier, reviewer, and team handoff flows.
Attribute feedbackPOST /v1/feedbackfeedback(), execution.feedbackFromOutcome()NoPrimaryTies outcome back to memory IDs exposed by a guide.
Measure effectPOST /v1/measuremeasure(), execution.measureRun()NoPrimaryBuilds effect reports and trace-derived skill candidate data.
Rehydrate evidencePOST /v1/rehydraterehydrate()Usually noStableExpands compact evidence pointers on demand. Feed only the selected evidence text back to the Agent when needed.
Resolve memory evidencePOST /v1/memory/resolveresolveMemory()Usually noStableSDK guideAgentContext() uses this to resolve inspect/rehydrate evidence.
Govern external memoryPOST /v1/memory/governgovernMemory(), governMem0SearchResults()Yes, via agent_context.prompt_textStableMemory Firewall path for Mem0, Zep, vector DB, markdown, logs, or custom candidates.
Forget or lifecycle-control memoryPOST /v1/forgetforget()NoOperator-onlySuppress, unsuppress, activate, or rehydrate with a reason.
Operator snapshotGET /v1/operator/snapshotoperatorSnapshot(), snapshot(), execution.snapshotRun()NoOperator-onlyRuntime state and replay summary for review.
Flight RecorderPOST /v1/audit/flight-recorderflightRecorder()NoOperator-onlyReplays what the Agent could see and what was blocked.
Decision tracePOST /v1/debug/memory-decision-traceNo primary SDK helperNoOperator-onlyDebug and dashboard surface.
Decision reportPOST /v1/audit/memory-decision-reportNo primary SDK helperNoOperator-onlyAudit report surface.
HealthGET /healthhealth()NoStableRuntime readiness and provider configuration.

SDK Surface

SDK method or helperCallsAgent-facingStatusUse when
createAionisClient()N/ANoPrimaryCreating a Runtime client.
observe()/v1/observeNoPrimaryYou already have a product observe payload.
remember()/v1/observeNoStableYou want to store an ordinary fact, preference, or note.
execution.observeStep()/v1/observeNoPrimaryYou are recording an execution step, branch, verifier result, or target files.
execution.handoff()/v1/observeNoPrimaryYou are recording cross-session or cross-agent continuation state.
guide()/v1/guideYes, nestedStableYou want raw HTTP guide response shape.
guideAgentContext()/v1/guide, /v1/memory/resolveYes, top-level agent_promptPrimaryYou want the easiest Agent-ready context for a Node/TypeScript host.
execution.guideForRole()/v1/guideYes, nestedStableYou want raw role-aware guide response shape.
execution.guideAgentContextForRole()/v1/guide, /v1/memory/resolveYes, top-level agent_promptPrimaryYou want Agent-ready context for planner, worker, verifier, reviewer, or team flows.
compileExecutionAgentContext()Local compile helperYes, top-level agent_promptAdvancedYou already have a guide object and need to render the SDK execution contract locally.
agentPromptFromGuide()Local extraction helperYes, stringAdvancedYou only need raw agent_context.prompt_text from a guide.
feedback()/v1/feedbackNoPrimaryYou already have feedback payload.
feedbackFromGuide()Local helperNoPrimaryYou need a feedback payload tied to a guide trace and used memory IDs.
execution.feedbackFromOutcome()/v1/feedbackNoPrimaryYou record an execution outcome and memory attribution together.
measure()/v1/measureNoPrimaryYou already have measure input.
execution.measureRun()/v1/measureNoPrimaryYou measure a run after guide and feedback.
measureInputFromGuideLoop()Local helperNoStableYou need a bounded measure request from a guide loop.
traceDerivedSkillCandidatesFromMeasure()Local helperNoStableYou want raw trace-derived skill candidates from a measure result.
traceDerivedSkillReviewItemsFromMeasure()Local helperNoStableYou want compact review items for operator review.
governMemory()/v1/memory/governYes, nestedStableYou pass external memory candidates into Aionis.
governMem0SearchResults()/v1/memory/governYes, nestedStableYou want a Mem0 drop-in governance boundary.
rehydrate()/v1/rehydrateNoStableYou need raw evidence behind a compact pointer.
resolveMemory()/v1/memory/resolveNoStableYou need to resolve an Aionis URI yourself.
forget()/v1/forgetNoOperator-onlyYou are changing lifecycle state with an audit reason.
snapshot()/v1/operator/snapshotNoOperator-onlyYou need a run or operator snapshot.
flightRecorder()/v1/audit/flight-recorderNoOperator-onlyYou need replay and audit evidence.

Integration Surface

IntegrationPackage or repoAgent-facing outputUses same context semanticsStatusNotes
SDK@aionis/sdkguideAgentContext().agent_promptYesPrimaryMain product path for Node and TypeScript hosts.
MCP@aionis/mcpaionis_context.agent_promptYesIntegrationTool bridge for MCP-capable coding Agents.
AIFS@aionis/aifs.aionis/guide.mdYesIntegrationFile surface for Agents that read workspace files.
Claude Codeaionis-claude-codeinjected hook contextYesIntegrationNative lifecycle hooks plus MCP setup.
HTTPRaw Runtime APIlower-level agent_context.prompt_textYesStableUse when you are outside TypeScript or need service boundaries; SDK AgentContext remains the recommended final prompt path.
Substrate@aionis/substrateNone as final promptNoAdvancedDurable evidence sidecar and recall candidate source. Runtime still governs guide output.
Manifest@aionis/manifestRuntime handoff evidenceNoAdvancedWorkflow and executable manifest layer, not a default Agent context renderer.
Dashboardaionis-dashboardNoneNoOperator-onlyRead-only trust window over snapshot, trace, measure, and flight recorder.

Setup Surface

Install commandWhat it doesStatus
npx aionis setupInstalls local Runtime, writes .env, installs dependencies, builds Runtime.Primary
npx aionis setup --profile full-localEnables existing AIFS setup guidance and the Zvec ANN candidate index.Primary local profile
npx aionis setup --with-zvec-annEnables Runtime Zvec ANN candidate index and verifies @zvec/zvec.Advanced
npx aionis setup --with-aifsPrints AIFS setup commands for the Agent workspace.Integration
npx aionis setup --with-claude-codeRuns Claude Code onboarding and hook installation.Integration
npx @aionis/aifs@latest initCreates the .aionis/ file surface in an Agent workspace.Integration
npx @aionis/mcp@latestStarts the MCP bridge against a running Runtime.Integration
Manual Substrate installInstalls and configures @aionis/substrate as a sidecar.Advanced

There is no separate Full Aionis Runtime. The local full profile composes existing pieces: Runtime, embeddings, AIFS setup guidance, and optional Zvec ANN. Claude Code hooks, MCP, and Substrate remain explicit integration choices.

Do Not Expose To Agents

SurfaceReason
memory_packetInternal product packet. Useful for audit and debugging, not prompt material.
guide_packetInternal guide packet. The Agent should receive compiled agent_context, not packet internals.
memory_decision_traceDebug surface. Can contain decision internals that should stay in host/operator logs.
memory_decision_auditOperator audit report.
memory_use_receiptUsually host/operator metadata. Pass only selected summaries if your host needs to explain attribution.
operator_snapshotRuntime state snapshot for operators and dashboards.
flight_recorderReplay report for audit, not next-turn instruction.
raw slots and raw payloadsRehydrate selectively and only when needed for exact evidence.

Explicit Non-goals

Do not buildWhy
A second SDK final-context methodguideAgentContext().agent_prompt already exists and is the primary SDK path.
A second final-Agent-context HTTP endpoint/v1/guide -> agent_context.prompt_text is the HTTP path; SDK adds agent_prompt for TypeScript hosts.
Substrate as Runtime replacementRuntime SQLite remains the Runtime fact source and admission authority.
Dashboard as Agent contextDashboard is read-only operator review.
Benchmark runner inside RuntimeEval workspaces validate Runtime; they do not define product behavior.
Manifest as default Agent contextManifest can create workflow evidence and handoffs; it is not the default context renderer.
Last updated on