Skip to main content

Agience Solution Codenames (End-State Taxonomy)

Status: Reference Date: 2026-04-01

This taxonomy is public architecture vocabulary. It explains the product roles Agience is organized around.

Eight codenames

CodenameDomainOne-liner
AstraIngestion, Validation & IndexingCapture and prepare incoming information
SageResearch & RetrievalDiscover sources, gather evidence, rank relevance
VersoReasoning & OrchestrationSynthesize, coordinate multi-step tasks, evaluate
AriaPresentation & InterfaceFormat, visualize, narrate — last mile to humans
IrisNetworking, Transport & RoutingConnect agents, services, and external systems
SeraphSecurity, Governance & TrustAccess control, audit, identity, policy compliance
MantleProvenance, Attribution & LineageVersion lineage, conflict detection, traceability
OphanFinance, Economic Logic & PerformanceAccounting, trade, resource allocation, budgeting, metrics

Pipeline position

Input: Astra — ingest, validate, normalize, index
Retrieval: Sage — search, retrieve, gather evidence
Reasoning: Verso — synthesize, reason, run operators
Output: Aria — format, visualize, present to humans

Cross-cutting:
Iris — networking, messaging, routing, remote execution
Seraph — security, identity, access control, policy
Mantle — provenance, lineage, attribution, coherence
Ophan — finance, accounting, budgets, performance metrics

Core codenames

Astra — ingestion, validation & indexing

Primary responsibilities:

  • Capture incoming information (files, URLs, streams, connectors)
  • Validate and normalize content before it enters the workspace
  • Deduplicate, classify, and index artifacts for retrieval
  • Register external connectors as Agience-side artifact records (data pull routes through the official vendor MCP server)
  • Telemetry collection from system inputs and activity streams

Typical MCP/tool surfaces:

  • ingest_file — create workspace artifacts from file URLs or raw text
  • validate_input, normalize_artifact, deduplicate, classify_content
  • connect_source, sync_source — connector artifact registration + sync (official-first)
  • list_streams, transcribe — live streaming integration
  • collect_telemetry

Sage — research & retrieval

Primary responsibilities:

  • Hybrid search across workspaces and collections (BM25 + semantic kNN)
  • Evidence-backed synthesis from retrieved sources
  • External retrieval connectors (Azure AI Search)
  • Research operations that produce reviewable outputs with provenance
  • Meeting/transcript insights

Typical MCP/tool surfaces:

  • search, get_artifact, browse_collections
  • search_azure, index_to_azure — optional Azure AI Search connector
  • research, cite_sources — multi-step retrieval + LLM synthesis (future)
  • ask, extract_information, generate_meeting_insights

Verso — reasoning & orchestration

Primary responsibilities:

  • Synthesize information from multiple sources via LLM
  • Execute defined multi-step operators
  • Chain tool calls sequentially across agents
  • Schedule deferred actions
  • Evaluate output quality and submit training feedback

Typical MCP/tool surfaces:

  • synthesize — LLM synthesis with artifact context
  • run_operator, chain_tasks, schedule_action
  • evaluate_output, submit_feedback

Artifact types owned:

  • application/vnd.agience.evaluation+json — evaluation result

Aria — presentation & interface

Primary responsibilities:

  • Format content for human consumption (markdown, HTML, plain text)
  • Render visualizations (charts, diagrams, tables)
  • Adapt tone and language for target audiences
  • Present workspace artifacts with appropriate layout
  • Generate natural-language narratives from structured data

Typical MCP/tool surfaces:

  • format_response, present_artifact
  • render_visualization, adapt_tone, narrate

Iris — networking, transport & routing

Primary responsibilities:

  • Platform-native messaging via channel adapters (Telegram, Slack, email)
  • Webhook ingestion from external systems
  • Service endpoint registration and routing
  • Shell execution in sandboxed environments
  • Secure tunnels from local host to platform
  • Proxy tool calls to registered external MCP servers (official-first)

Typical MCP/tool surfaces:

  • send_message, get_messages, list_channels
  • create_webhook, health_check, list_connections
  • register_endpoint, route_request
  • exec_shell, tunnel, proxy_tool

Seraph — security, governance & trust

Primary responsibilities:

  • Access control (grant/revoke collection access)
  • Audit trails (who accessed what, when)
  • Identity verification (JWT/API key claims)
  • API key lifecycle (rotation, scoping)
  • Cryptographic signing of artifacts
  • Policy enforcement and compliance checking

Typical MCP/tool surfaces:

  • audit_access, list_audit_events
  • check_permissions, grant_access, revoke_access
  • rotate_api_key, verify_token
  • sign_artifact, enforce_policy, list_policies, check_compliance

Mantle — provenance, attribution & lineage

Primary responsibilities:

  • Trace artifact version lineage and source attribution
  • Detect conflicting claims across workspace or collection
  • Validate workspaces against contract artifacts
  • Propose merges of divergent artifact versions
  • Graph traversal across knowledge relationships
  • Human-in-the-loop approval gates

Typical MCP/tool surfaces:

  • check_provenance, attribute_source
  • detect_conflicts, check_coherence
  • apply_contract, suggest_merge
  • traverse_graph

Ophan — finance, economic logic & performance

Primary responsibilities:

  • Value transfers (crypto on-chain, fiat bank/wire)
  • Double-entry ledger recording and reconciliation
  • Invoice creation and payment application
  • Market data (price feeds, OHLCV, portfolio)
  • Resource usage tracking and budget calculation
  • System performance and efficiency metrics

Credentials model: BYOK — exchange/bank keys stored as encrypted user secrets, retrieved per-request from /secrets. Never in server env vars.

Typical MCP/tool surfaces:

  • send_payment, record_transaction, get_transaction, list_transactions
  • fetch_statement, get_balance, reconcile_account
  • create_invoice, apply_payment, run_report
  • get_price, get_market_data, track_wallet, get_portfolio, calculate_pnl
  • track_resource_usage, get_metrics, calculate_budget

Pillar mapping

PillarPrimary codenameNotes
Ingestion & IndexingAstraCapture, validate, normalize, index
Research & RetrievalSageHybrid search, evidence synthesis, meeting insights
Reasoning & OrchestrationVersoSynthesis, multi-step operators, evaluation
Presentation & InterfaceAriaFormat, visualize, narrate to humans
Networking & InfrastructureIrisMessaging, routing, tunnels, proxy
Security & GovernanceSeraphAccess control, audit, identity, policy
Provenance & LineageMantleAttribution, conflict detection, contracts
Finance & PerformanceOphanAccounting, trade, metrics, budgets

Notes

  • Running operators and recording provenance are core Agience framework capabilities (not a separate pillar). Provenance travels with every artifact's context — no separate receipt object is created.
  • Official-first: Agience never duplicates what an official vendor MCP server provides. External servers (GitHub, AWS, filesystem, Slack, etc.) are registered as vnd.agience.mcp-server+json artifacts. Iris's proxy_tool and Astra's connect_source/sync_source are the integration points — they route through official servers, not around them.