START HERE. Bootstrap a governed constitutional session and bind actor identity. Select when: starting any new conversation, resuming a previous session, or before any other arif_* tool call. Without a session_id, no governed action is possible. Returns: session_id, authority level (SEAL/OBSERVE_ONLY/FULL), floor status, next_tool recommendation, and constitution hash. Modes: init (full, ~60s) | light (<1s, tool pointers only) | resume | validate | epoch_open | epoch_seal | ping. Do NOT select when: you already have a live session_id and verified authority (use arif_triage instead), or the request is a pure factual question needing no session.
Parameters (26)
modestring
actor_idany
ack_irreversibleboolean
session_idany
epoch_idany
previous_session_hashany
declared_model_keyany
actor_signatureany
nonceany
counterpartyany
contextany
evidenceany
toolingany
agent_policyany
intentany
requested_authoritystring
verboseany
idempotency_keyany
trace_idany
caller_actor_idany
executor_actor_idany
sovereign_idany
delegation_modeany
payloadany
_envelopeany
client_capabilitiesany
arif_observe
Ground your next decision in reality. Select when: you need external evidence (web search, URL fetch, system vitals), need to map a repository structure, or need to compute entropy from a dataset. The gradient pulling toward this tool is unresolved factual uncertainty — if the answer requires data not in your context, choose observe. Modes: search | ingest | compass | atlas | entropy_dS | vitals | repo_map | hybrid_discovery. Returns: search results, ingested content, system vitals, entropy scores, or repo map. Do NOT select when: the user's intent is purely reasoning/analysis (choose arif_think instead), or the evidence is already in context.
Parameters (8)
modestring
queryany
session_idany
actor_idany
urlany
layersany
result_limitinteger
_envelopeany
arif_think
Reason, plan, reflect, or critique — the cognitive engine. Select when: you need to decompose a complex problem, generate a plan, evaluate hypotheses, verify a conclusion, critique a proposal, or synthesize evidence into a recommendation. The gradient pulling toward think is cognitive overload — when the next step requires structured reasoning rather than observation or action. Modes: reason (decompose) | reflect (evaluate session) | verify (check conclusion) | critique (assess risks) | plan (generate DAG) | plan_review | plan_approve | refactor_plan | metabolize (synthesize) | axioms. Returns: structured reasoning output with epistemic labels (OBS/DER/INT/SPEC), facts, inferences, unknowns, confidence bands, and next_safe_action. Do NOT select when: the question is factual and requires external data (select arif_observe first), or when immediate action is needed (select arif_act).
Parameters (7)
modestring
queryany
session_idany
actor_idany
plan_idany
witness_typestring
_envelopeany
arif_route
Canonical intent router. Select when: you know what you want but not which tool or organ to call. Routes natural-language intent to the correct federation organ (GEOX, WEALTH, WELL, A-FORGE) or kernel tool. Optionally accepts organ_tool to bridge-call directly — bypassing the routing decision. Returns: organ, port, tool_prefix, suggested_tools. Do NOT select when: you already know the exact tool to call.
Parameters (8)
intentstringrequired
Natural-language description of what the user wants.
e.g. "interpret this seismic section", "assess portfolio risk"
organany
Optional explicit organ override. If provided, intent matching
is skipped and this organ is used directly.
taskany
Alias for intent (backward compat).
actor_idany
Calling actor.
session_idany
Governing session.
organ_toolany
The tool name on the target organ to call.
If absent, returns routing decision only (no bridge call).
argumentsany
Arguments to pass to organ_tool.
_envelopeany
arif_judge
Render final constitutional verdict — the arbitration gate. Select when: a decision is ready for binding judgment. You have gathered evidence (observe), reasoned about it (think), and the decision requires floor compliance verification. The gradient toward judge is completion of the evidence-to-plan pipeline. Returns: SEAL (approved) | HOLD (revise) | SABAR (wait) | VOID (rejected) — with violated floors, evidence receipts, and authority verification. Modes: judge | compare | history | explain | floor_status | witness_consensus. REQUIRES: actor, intent, domain, reversibility_level, blast_radius — all mandatory. Do NOT select when: evidence is incomplete (choose observe first), plan is not ready (choose think first), or the action is reversible and low-risk (advisory mode is sufficient).
Parameters (12)
actorstringrequired
intentstringrequired
requested_capabilitystringrequired
domainstringrequired
reversibility_levelstringrequired
blast_radiusstringrequired
epistemic_statestring
evidenceany
authority_tokenany
_envelopeany
actor_idany
session_idany
arif_act
Execute an approved action — the 900 execution gate. HARD REQUIREMENT: Valid prior SEAL from arif_judge + arif_seal is mandatory. Select only when: you have seal_verdict_id AND approved_action_hash from a completed judge→seal pipeline. The gradient toward act is complete authorization — all evidence gathered, plan reasoned, risk critiqued, judge approved, seal written. Without seal_verdict_id + approved_action_hash, this tool returns 888_HOLD structurally. Routes through A2ASealVerifier for cryptographic verification before execution. Do NOT select when: you are still in planning/critique phase, or no prior SEAL exists. This is the LAST tool in the constitutional pipeline — after it, seal the result.
Parameters (6)
seal_verdict_idstringrequired
approved_action_hashstringrequired
manifestany
actor_idany
session_idany
_envelopeany
arif_seal
Append a verdict or outcome to the immutable VAULT999 ledger — irreversible. Select when: a 888_JUDGE SEAL verdict exists and needs permanent cryptographic anchoring, or an execution result needs audit trail. The gradient toward seal is finality — this record can never be deleted. Modes: seal (append record) | verify (validate chain integrity) | chain (show hash chain) | list (enumerate seals) | dry_run (preview) | seal_card | render. ack_irreversible=True is required for seal mode. Do NOT select when: the verdict is HOLD/SABAR/VOID (seal only SEAL outcomes), or testing (use dry_run mode instead).
Parameters (12)
modestring
payloadstring
session_idany
ack_irreversibleboolean
actor_idany
actor_signatureany
nonceany
constitutional_chain_idany
judge_state_hashany
witness_typestring
drift_eventsany
_envelopeany
arif_conformance_report
Full conformance spine proof machine. Runs 9 live checks against the running arifOS kernel: arifos_alive, mcp_initialize, protocol_version, schema_echo_stable, session_starts, authority_checked, hold_blocks_mutation, vault_replay, cooling_ledger. Returns structured PASS/FAIL report with evidence per check. No session required. Read-only. Zero side effects. Use when: you need to prove arifOS is a substrate, not just a narrative.
Parameters (7)
payloadany
_envelopeany
client_capabilitiesany
actor_idany
session_idany
modestring
limitinteger
arif_canary
Unified transport diagnostic probe. One tool, six modes. Use for liveness checks, protocol version verification, schema round-trip testing, transport detail dumps, MCP handshake tests, and full conformance spine. Modes: ping | schema_echo | version_echo | transport_echo | initialize_probe | conformance_report
Parameters (7)
modestring
payloadany
_envelopeany
client_capabilitiesany
session_idany
actor_idany
include_constitutionboolean
arif_triage
Session status, priority queue, and preflight checks. Modes: status (active session count + stage) | preflight (pre-session safety probe) | triage (priority assessment). Core immune function — run before arif_init when unsure of session state.
Parameters (6)
modestring
"status" | "preflight" | "triage"
session_idany
Optional session to query
stageany
Stage hint (used if session_id not provided)
actor_idany
Calling actor
priorityany
Task priority hint for triage mode
_envelopeany
arif_compose
Governed response composition — formats final output for Arif with citations and tone calibration. Call as the LAST step before presenting results. Modes: compose | summarize | cite | tone_shift | style | format | nudge | repo_answer.
Parameters (9)
modestring
messageany
styleany
citationsany
session_idany
actor_idany
ai_involvementstring
languagestring
_envelopeany
forge_query
forge_* tools have moved to A-FORGE MCP. Connect to forge.arif-fazil.com/mcp for engineering tools. This arifOS endpoint will be removed 2026-07-15.
Parameters (6)
manifeststring
querystring
cwdstring
session_idany
actor_idany
_envelopeany
forge_plan
forge_* tools have moved to A-FORGE MCP. Connect to forge.arif-fazil.com/mcp for engineering tools. This arifOS endpoint will be removed 2026-07-15.
Parameters (5)
goalstring
workspacestring
session_idany
actor_idany
_envelopeany
forge_dry_run
forge_* tools have moved to A-FORGE MCP. Connect to forge.arif-fazil.com/mcp for engineering tools. This arifOS endpoint will be removed 2026-07-15.
Parameters (6)
plan_idstring
manifeststring
cwdstring
session_idany
actor_idany
_envelopeany
forge_plan_and_simulate
forge_* tools have moved to A-FORGE MCP. Connect to forge.arif-fazil.com/mcp for engineering tools. This arifOS endpoint will be removed 2026-07-15.
Parameters (5)
intentstring
contextany
risk_tierstring
force_simulationboolean
_envelopeany
arif_detect_institutional_shadow_drift
Detect when a sovereign institution's observed functions have outgrown its declared name (GENESIS/006 Petronas Paradox). Returns drift_score, sovereignty_score, risk_class, verdict, and cross-node routing hints. Use before high-blast-radius deals involving national institutions.
Parameters (9)
institution_namestringrequired
declared_functionstringrequired
observed_functionsarrayrequired
declared_function_keywordsany
last_shadow_updateany
capital_exposure_myrany
actor_idany
session_idany
_envelopeany
arif_detect_narrative_tension
Detect paradox tension, power asymmetry, and implicit frames in news articles or institutional text. Returns a FrameGraph with actors, claims, tensions, and a kernel verdict. Golden case: Putra Heights Kosmo 2026-06-12.
Parameters (13)
titlestringrequired
Article headline.
textstringrequired
Full article text.
article_idany
Optional stable identifier; triggers golden-case lookup for known articles.
sourcestring
authorany
published_atany
urlany
tagsany
Subject tags.
public_intereststring
LOW | MEDIUM | HIGH | CRITICAL
actor_hintsany
Optional hints for actor extraction.
actor_idany
session_idany
_envelopeany
arif_os_attest
arifOS organ.attest(): live self-attestation of the constitutional kernel. Returns constitution_hash, schema_hash, tool_surface, health, and active lease state. Required before any kernel-grade federation call.
Parameters (3)
actor_idany
session_idany
_envelopeany
arif_organ_attest
Probe and attest a federation organ (GEOX, WEALTH, WELL). Returns organ heartbeat, schema hash, tool count, and kernel envelope.
Parameters (4)
organ_idstringrequired
actor_idany
session_idany
_envelopeany
arif_organ_attest_all
Attest arifOS plus all federation organs in one call. Returns per-organ heartbeat and a degraded-organ list.
Parameters (3)
actor_idany
session_idany
_envelopeany
arif_lease_issue
Issue, inspect, or revoke a bounded authority lease. Leases scope organ/agent tool access and action class.
Parameters (9)
organ_idstringrequired
actor_idanyrequired
scopearrayrequired
max_action_classstring
ttl_secondsinteger
forbiddenany
session_idany
max_usesany
_envelopeany
arif_lease_inspect
Issue, inspect, or revoke a bounded authority lease. Leases scope organ/agent tool access and action class.
Parameters (5)
lease_idany
organ_idany
actor_idany
session_idany
_envelopeany
arif_lease_revoke
Issue, inspect, or revoke a bounded authority lease. Leases scope organ/agent tool access and action class.
Parameters (5)
lease_idstringrequired
reasonstring
session_idany
actor_idany
_envelopeany
arif_heartbeat
Record or query federation heartbeats. Returns liveness verdict for known organs.
Parameters (4)
organ_idany
session_idany
actor_idany
_envelopeany
arif_peer_contract_validate
Peer Federation Contract v1 — validate.
Parameters (5)
contractany
contract_urlany
actor_idany
session_idany
_envelopeany
arif_peer_contract_attest
Peer Federation Contract v1 — attest.
Parameters (4)
organstring
actor_idany
session_idany
_envelopeany
arif_peer_contract_forbid
Peer Federation Contract v1 — forbid.
Parameters (5)
organstringrequired
reasonstring
actor_idany
session_idany
_envelopeany
arif_resolve_tool
Resolve a tool name or alias to the canonical arifOS tool name. Use when you have a tool name but aren't sure if it's the canonical name. Returns: canonical_name, aliases, callable, schema_valid, authority_class, use_when, examples. Only `name` is used — all other fields are absorbed and ignored.
Parameters (4)
namestringrequired
_envelopeany
actor_idany
session_idany
arif_get_affordance
Return the complete constitutional affordance contract for a tool name. Use this to decide 'why this tool', 'why not others', risk, agency level, and expected metacognition shape. Preferred over guessing from name alone.
Parameters (4)
namestringrequired
_envelopeany
actor_idstring
session_idstring
arif_retrieve_tools
BM25 lexical tool retrieval across the arifOS federation tool catalog. Schema-aware text projection: tool name + description + parameter names + enum values, JSON syntax stripped. k1=0.9 b=0.4 tuned for short tool descriptions (Ratel ADR-0004). Cross-organ by default; set organ= to restrict to one organ. F2 TRUTH: BM25 scores are TERM FREQUENCY lexical match, NOT confidence. F7 HUMILITY: Scores measure keyword overlap, not semantic relevance. Use as DISCOVERY hints to find the right tool, then verify with arif_resolve_tool. Pair with arif_route for the full pipeline: intent → organ → BM25 rank → select. Parameters: query (natural language intent), organ (optional, restrict to organ), top_k (1-20, default 5), include_scores (default true).
Parameters (7)
querystringrequired
organany
top_kinteger
include_scoresboolean
_envelopeany
actor_idany
session_idany
hermes_system_status
HERMES_SYSTEM_STATUS: Return current federation state snapshot.
Modes:
brief — organ health + latest event count (default)
full — brief + VAULT999 seal count + memory stats
organs — organ health only
events — NATS governance events only
F2: All organ health data from live TCP probes.
Parameters (4)
modestring
actor_idany
_envelopeany
session_idany
hermes_vault_query
ARIF_VAULT_QUERY: Query VAULT999 audit ledger.
Modes:
recent — return N most recent VAULT entries (default)
search — search VAULT entries by keyword
organ — filter entries by organ name
date — filter entries by date prefix (YYYY-MM-DD)
Args:
query — search term or organ name or date (depends on mode)
limit — max entries to return (default 10, max 50)
actor_id— acting agent identity
F2: All data direct from VAULT999 filesystem.
Cycle 3 fix (2026-06-21): the conformance spine's `vault_replay` check
needs each entry to expose `ts`/`timestamp` (alias of mtime), and the
response itself to expose `status` + `chain_ok` INSIDE the result dict
(because `_extract_tool_result` strips the outer `status` key when
returning parsed["result"]). Without these the kernel could not prove
it can read its own sealed past — the load-bearing property of a
substrate. Now it can.
Parameters (6)
querystring
modestring
limitinteger
actor_idany
_envelopeany
session_idany
arif_vault_query
ARIF_VAULT_QUERY: Query VAULT999 audit ledger.
Modes:
recent — return N most recent VAULT entries (default)
search — search VAULT entries by keyword
organ — filter entries by organ name
date — filter entries by date prefix (YYYY-MM-DD)
Args:
query — search term or organ name or date (depends on mode)
limit — max entries to return (default 10, max 50)
actor_id— acting agent identity
F2: All data direct from VAULT999 filesystem.
Cycle 3 fix (2026-06-21): the conformance spine's `vault_replay` check
needs each entry to expose `ts`/`timestamp` (alias of mtime), and the
response itself to expose `status` + `chain_ok` INSIDE the result dict
(because `_extract_tool_result` strips the outer `status` key when
returning parsed["result"]). Without these the kernel could not prove
it can read its own sealed past — the load-bearing property of a
substrate. Now it can.
Parameters (6)
querystring
modestring
limitinteger
actor_idany
_envelopeany
session_idany
hermes_epistemic_check
HERMES_EPISTEMIC_CHECK: Pre-flight epistemic confidence check for a CLAIM.
Evaluates a claim against available evidence to produce a
CONFIDENCE_LEVEL, GAPS, and RECOMMENDATION before the claim
enters the metabolic pipeline.
Modes:
quick — fast heuristic check (default)
vault — cross-reference against VAULT999 entries
full — vault + memory + organ consensus (slowest)
Args:
claim — the claim to evaluate
evidence_context — optional context describing available evidence
actor_id — acting agent identity
F2: Returns CONFIDENCE label with explicit gap analysis.
F7: Never returns overconfident labels.
Parameters (6)
claimstring
modestring
evidence_contextstring
actor_idany
_envelopeany
session_idany
hermes_fact_check
HERMES_FACT_CHECK: Verify a CLAIM against evidence sources.
Uses web_search + VAULT999 + available tools to produce a structured
verdict about a factual claim.
Modes:
quick — local heuristic + VAULT999 check only (default)
web — quick + web_search for external corroboration
deep — web + VAULT999 + memory cross-reference
Returns:
verdict: CONFIRMED | REFUTED | MIXED | UNKNOWN
confidence: 0.0–1.0
evidence: list of sources found
gaps: list of missing evidence
F2: Returns structured evidence, not narrative.
F7: Never overclaims confidence.
Parameters (8)
claimstring
modestring
required_confidencenumber
time_sensitiveboolean
evidence_contextstring
actor_idany
_envelopeany
session_idany
hermes_cross_verify
HERMES_CROSS_VERIFY: Submit a CLAIM for cross-agent verification.
Delegates fact-check to a second agent (OpenCode by default) for
independent verification. Returns structured result only.
Args:
claim — the claim to verify
target — verification target: "opencode" (default)
time_budget_seconds — max time to wait for verification (default 60)
Returns:
status: OK | ERROR | TIMEOUT
verdict: VERIFIED | CONTRADICTED | INSUFFICIENT
evidence: list of evidence discovered
confidence: 0.0–1.0
F2 + F3: Cross-agent witness methodology.
Parameters (6)
claimstring
targetstring
time_budget_secondsinteger
actor_idany
_envelopeany
session_idany
hermes_plan_review
HERMES_PLAN_REVIEW: Review a multi-step plan for safety and completeness.
Checks:
- Missing safety/verify steps
- F1-F13 floor violations
- Unclear success criteria
- Risk assessment per step
- E7 autonomy ceiling
Args:
plan — JSON string or numbered list of steps
goal — what the plan aims to achieve
mode — quick | full
Returns:
status: OK | WARN | BLOCK
issues: list of potential problems
recommendations: list of suggested changes
Parameters (6)
planstring
goalstring
modestring
actor_idany
_envelopeany
session_idany
hermes_memory_steward
HERMES_MEMORY_STEWARD: Classify content for memory storage.
Evaluates content and recommends where it should go in the
memory hierarchy:
- STORE_IN_VAULT — high-importance, permanent audit trail
- STORE_IN_GRAPHITI — relational knowledge, edges between concepts
- STORE_IN_MEMORY — operational context, medium-term
- DISCARD — ephemeral, not worth storing
- TODO_FOR_ARIF — requires human review
Args:
content — the content to classify
importance — low | medium | high | critical
mode — classify | compact
Returns:
recommendation: storage target
justification: why this classification
Parameters (6)
contentstring
importancestring
modestring
actor_idany
_envelopeany
session_idany
arif_wiki_ingest
Index a local repository into the wiki. Run this FIRST before map, search, or ask.
Parameters (5)
repo_pathanyrequired
scope_namestring
include_globsany
exclude_globsany
write_wikiboolean
arif_wiki_map
Get a structural map of an indexed repository. Run arif_wiki_ingest first.
Parameters (2)
repo_pathanyrequired
max_depthinteger
arif_wiki_search
Search the wiki index for scored evidence chunks. Run arif_wiki_ingest first.
Parameters (3)
repo_pathanyrequired
querystringrequired
top_kinteger
arif_wiki_ask
Ask a natural-language question over the wiki index. Run arif_wiki_ingest first.
Parameters (3)
repo_pathanyrequired
questionstringrequired
top_kinteger
arif_stack_health_probe
Probe federation health: arifOS, organs, vault, and model registry.
Parameters (3)
session_idany
Governed session ID for audit trace.
actor_idany
Sovereign actor identifier.
include_servicesany
Subset of services to probe. Default = all.
arif_organ_consensus
Request cross-organ consensus from WELL, WEALTH, and GEOX.
Parameters (4)
action_descriptionstring
Free-text description of the proposed action.
organsany
Subset of organs to consult (default: all three).
session_idany
Governed session ID for audit trace.
actor_idany
Sovereign actor identifier.
arif_scan_local_instructions
Scan files for hidden instructions or governance violations.
Parameters (3)
root_dirany
session_idany
actor_idany
arif_session_budget
Check token, compute, and financial budget for this session.
Parameters (10)
modestring
decision_classstring
delta_snumber
costnumber
risknumber
action_descriptionstring
objective_idany
ceiling_overrideany
session_idany
actor_idany
arif_floor_status
Report the current state of all 13 constitutional floors.
Parameters (2)
session_idany
actor_idany
mcp_drift_check
Compare live tool registry against canonical manifest for drift.
█████╗ ██████╗ ██╗███████╗ ██████╗ ███████╗
██╔══██╗██╔══██╗██║██╔════╝██╔═══██╗██╔════╝
███████║██████╔╝██║█████╗ ██║ ██║███████╗
██╔══██║██╔══██╗██║██╔══╝ ██║ ██║╚════██║
██║ ██║██║ ██║██║██║ ╚██████╔╝███████║
╚═╝ ╚═╝╚═╝ ╚═╝╚═╝╚═╝ ╚═════╝ ╚══════╝
Constitutional AI Governance Kernel
─────────────────────────────────────────
Not a chatbot. Not a model wrapper. The LAW.
DITEMPA BUKAN DIBERI — "Forged, Not Given."
</div>
1. What Is arifOS?
arifOS is a constitutional governance kernel that sits between AI agents and their tools, enforcing 13 floors before any irreversible action.
The law layer — decides what agents must NOT do, so they can be trusted with what they CAN do
An MCP server — 10 public verbs. Agents call arif_init → arif_judge → arif_seal
A federation hub — 7 organs (arifOS, A-FORGE, GEOX, WEALTH, WELL, AAA, VAULT999) under one contract
An immutable ledger — VAULT999: append-only, hash-chained. Every decision sealed forever
Built for one sovereign — Muhammad Arif bin Fazil. F13 veto is absolute
What it is NOT: an AI model, a chatbot, a startup, LangChain/CrewAI/AutoGen, or a replacement for human judgment.
APEX STACK Bridge
APEX THEORY defines the constitutional dynamics of governed intelligence through ΔΩΨ. arifOS compiles those dynamics into an AGI substrate kernel. AAA renders the substrate as visible ASI civilization state. A-FORGE gives the system governed hands. GEOX, WEALTH, and WELL anchor those hands to earth, capital, and human reality. VAULT999 preserves consequence. Arif/F13 remains the sovereign witness and final veto.
arifOS must never: replace human judgment, self-authorize a SEAL, skip 888_JUDGE, or issue verdicts without a constitutional chain.
