OMK

Open Multi-Agent Kit
Scope the work. Route the right agents. Verify every release.

Provider-neutral coding-agent CLI and multi-agent control plane for Codex, Claude Code, OpenCode, and local models.

Quick start · Control loop · Safety boundary
Packages · Documentation

---

Why OMK

Coding agents can produce code quickly. They can also overlap work, lose state, route to the wrong model, and claim completion before the build is green. OMK adds a control plane around that work.

Problem	OMK invariant	Inspectable output
Parallel agents overwrite each other	Owned paths and resource claims bound every lane	DAG and workspace state
A model says “done” too early	Acceptance predicates require fresh evidence	Commands, exits, and receipts
A provider or model changes	Routing stays separate from the execution contract	Provider-attributed attempts
A session stops midway	Replayable state supports session recovery	Ledger, repair plan, durable goal

OMK is for engineering work that needs a checkable result, not a convincing chat response.

Quick start

npm install -g open-multi-agent-kit --ignore-scripts
omk --version
omk

Without a global install:

npx --ignore-scripts open-multi-agent-kit

Requirements: Node.js 22.19 or newer. The published CLI package is open-multi-agent-kit.

Control loop

1. Scope — turn a goal into a bounded DAG with owned paths, ordered waves, resource claims, and acceptance predicates.
2. Route — select models, agent skills, MCP tools, and extensions for the job without changing the evidence contract.
3. Verify — run the declared build, type, test, audit, and release gates. Required red predicates block completion.
4. Replay — preserve receipts, repair interrupted sessions, and continue durable goals from explicit reducer state.

The animation changes once every 1.5 seconds and contains no flashing. The four steps above are the complete text alternative.

OMK//CONTROL

The default operator surface shows active work, routing, context, MCP state, and verification signals in one terminal UI.

The header reads omk v<package.version> · OMK//CONTROL; the installed package version is the source of truth.

What ships

Multi-agent execution

• Bounded DAG parallel agents with deterministic ready-lane ordering.
• Per-lane owned paths and resource claims.
• Explicit cancellation, timeout, and retry settlement.
• Durable goals and checkpointed continuation across bounded rounds.

Evidence and recovery

• Acceptance predicates backed by fresh command evidence.
• Versioned observations, evaluation results, and runtime decisions.
• Replay ledgers, receipts, session repair, and SDK session inspection.
• Advisory judging that cannot replace required deterministic gates.

Routing and extensibility

• Provider-neutral model registry through omk-ai.
• Agent skills loaded on demand instead of dumped into every prompt.
• MCP runtime client with lazy stdio server startup and registered tools.
• Extensions for tools, commands, events, providers, themes, and UI surfaces.

See providers, MCP, skills, and extensions.

Verification boundary

AgentSession built-in local bash uses OS sandbox enforcement by default: sandbox-exec on macOS and bwrap plus unprivileged user namespaces on Linux. Local shell spawns restrict writes to the workspace and OS temporary directory, disable network access, and fail closed with sandbox.backend_missing when an enforcement backend is unavailable.

This is not read-confidentiality or whole-process containment. Other file tools, extension and custom-tool code, injected or remote BashOperations, and the OMK process keep the permissions of the process running them. Use containerization when the boundary must cover more than built-in local bash. A run without required evidence remains UNVERIFIED.

Providers

OMK keeps routing separate from control and evidence. Codex, Claude Code, OpenCode Zen/Go, Kimi, GLM/ZAI, native xAI/Grok, NVIDIA NIM, and local providers can participate through omk-ai while the run contract stays stable.

Native xai keeps subscription OAuth and XAI_API_KEY billing separate. See provider setup, provider resilience, and Grok integration.

Published packages

Package	Purpose
open-multi-agent-kit	Interactive coding-agent CLI and control plane
omk-agent-core	Agent runtime, tool execution, and DAG scheduling
omk-ai	Unified multi-provider LLM API
omk-protocol	Versioned run contracts and semantic reducers
omk-adaptorch-wpl	Work Packet Loop runtime
omk-book-to-skill	Optional document-to-skill compiler
omk-tui	Differential-rendered terminal UI library

npm install omk-agent-core
npm install omk-ai
npm install omk-protocol
omk install npm:omk-book-to-skill@0.96.2
npm install omk-tui

OMK + AdaptOrch

OMK is the local, MIT-licensed control plane. AdaptOrch is a separate, proprietary hosted patch-evidence service. It requires its own account and is not part of this repository or the omk-adaptorch-wpl package.

Review AdaptOrch plans →

Documentation

• Documentation index
• Usage
• Providers and models
• Automation and SDK
• Run protocol
• Sessions and recovery
• Security
• Containerization
• Public skill catalog
• Changelog
• Release notes for v0.96.2

Development

npm ci --ignore-scripts
npm run build
npm run check
npm test
npm run release:local

Direct dependencies are pinned, CI installs with --ignore-scripts, and the published CLI includes a generated npm-shrinkwrap.json. Read CONTRIBUTING.md and the development guide before sending a change.

FAQ

Is OMK a coding agent or an orchestrator?

Both. open-multi-agent-kit is an interactive coding-agent CLI; OMK//CONTROL adds bounded orchestration, routing, evidence gates, and recovery.

Does OMK require one specific model provider?

No. Providers can change while the execution and evidence contracts remain stable. Provider-specific capabilities still vary and are documented explicitly.

How does OMK decide that work is complete?

Declared acceptance predicates must pass with fresh command evidence. Chat text, a reviewer opinion, or stale output cannot replace a required gate.

Can OMK recover an interrupted run?

Yes. Replay state, receipts, durable goals, and session repair preserve enough structure for bounded recovery instead of silently starting over.

Recent releases

Release v0.96.2

Added

• Added resource-aware host snapshots, admission decisions, generation-safe per-run tool-cap leases, workload classification, weighted FIFO permits, and omk doctor resources / /resource operator surfaces.
• Added internal Vitest, Jest, workspace, and Go shard planners plus a journaled executor with corruption quarantine, completed-shard resume, admission-aware execution, and aggregate workload_shard_result.v1 evidence. Automatic session-command sharding is not enabled.
• Added an internal subagent lane launcher that enforces parent admission width and shares its permit pool; live child-launch wiring is not enabled.
• Added exactly-once prompt_settled and an opt-in completion sound.
• Added a local-only resource observation journal at .omk/runs/<promptRunId>/resource-observations.jsonl, recording bounded probe health, admission caps, classification, permit waits, settlement, and sound outcomes without raw host measurements.
• The QWEN TOKEN PLAN status entry now uses the official QwenCloud management CLI to show the seven-day usage window and reset time.

Changed

• The session-termination classifier now accepts resource.* causes for memory, disk, CPU, heap, unavailable probes, and permit queue overflow. Only CPU pressure can qualify for automatic retry; live resource gates currently return bounded block results.
• resourceGovernor.mode: "observe" is the default and records decisions without enforcing caps. Both "observe" and "off" preserve v0.96.1 scheduling behavior.

Fixed

• Top-level omk --help now lists omk doctor resources [--json].
• Removed development-only prepare and postinstall hooks from the published CLI manifest; default npm installs no longer call an unshipped workspace-linking script.
• Bound resource observations and completion-sound results to their originating prompt journals, including consecutive fast observe-mode runs.
• Capped workspace shard plans at 16 while preserving every workspace through deterministic chunks.
• Ensured prompt_settled consistently follows agent_end as the final run event.

Security

• Qwen quota discovery never sends an inference API key to a management endpoint, never reads browser cookies, and passes only non-secret process context to the QwenCloud child CLI.
• Removed the shipped subagent example's offensive-jailbreak skill route.

Docs

• Documented an attributable AdaptOrch.com link for evaluating a separate hosted patch-evidence service; AdaptOrch remains distinct from the MIT-licensed OMK packages.

Release notes live in RELEASE_NOTES_v0.96.2.md.

Release v0.96.1

Added

• Added built-in harness loop extensions: identical-loop detection, compaction tool-pair repair, Kimi/K3/GLM/Grok/Claude prompt presets, and /goal with automatic continuation plus explicit pause, resume, evidence-gated complete, and clear lifecycle commands.
• Added pass-gated advisory best-of-N selection with strict weighted judge responses, forced-redacted candidate material, deterministic fallback, evaluation-bound request digests, and an explicit ModelRegistry-backed LLM adapter.
• Added digest-bound Goal / Core / Verified / Open / Next seam checkpoints to the existing durable-goal journal, including /goal checkpoint <json> and checkpoint-aware continuation.
• Added omk sdk session status|tail|inspect|send for external session controllers. Ambiguous selectors now fail closed, writes require exact IDs, active owners block concurrent access, and credential-shaped transcript text is redacted from output.
• Legacy read mode (OMK_OMP_SEAMS=0) now reports a private 0700 temporary spill directory with an exclusive 0600 file for recoverable line or byte truncation; it never writes beside or through the source path. A first line that alone exceeds the byte cap remains preview-only.

Changed

• Sandbox backend probing is now cached per local bash operations instance. All enforce-mode fallback verdicts share the concrete missing-backend diagnosis (bwrap, user namespaces, sandbox-exec, or unsupported platform). Policy semantics are unchanged.
• Extracted the session system-prompt assembly from AgentSession._rebuildSystemPrompt into the pure assembleSessionSystemPrompt module (core/session-system-prompt.ts). Provider playbook resolution stays at the call site; the assembly is now directly testable.
• Extracted the retry/failover decisions from AgentSession._isRetryableError and _prepareRetry into the pure core/provider-retry.ts module (isRetryableAssistantError, nextRetryAttempt, computeRetryDelayMs). Retry ordering, backoff, and failover semantics are unchanged.
• Extracted the compaction gates from AgentSession._checkCompaction into the pure core/compaction-gate.ts module (shouldSkipCompactionCheck, isSessionModelOverflow). Gate ordering and staleness semantics are unchanged.
• Extracted the failover trigger and refused-model bookkeeping from AgentSession._maybeFailoverFromSafetyStop into core/provider-retry.ts (isFailoverTriggerError, failoverModelKey). Chain ordering and blacklist semantics are unchanged.
• Extracted the context-budget arithmetic from AgentSession._getContextBudgetOptions into the pure core/prompt-budget.ts module (computePromptTokenBudget, computeResponseReserveTokens). Env parsing stays at the call site; budget values are unchanged.
• Extracted the prompt-cache key transition classification from AgentSession._recordPromptCachePlan into the pure core/prompt-cache.ts module (classifyPromptCacheTransition). Counter and break-reason semantics are unchanged.
• Grok 4.5 / 4.3 now expose /think max and ultra in the selector. Those aliases still send xAI reasoning_effort: "high" because those models have no upstream xhigh/max tier.
• Native xAI SuperGrok usage now polls GET https://cli-chat-proxy.grok.com/v1/billing?format=credits and shows the weekly SuperGrok pool from creditUsagePercent. Stale grok-oauth-proxy credentials are dropped from /login and /logout.

Fixed

• Durable-goal continuation now records and skips an unavailable WSL/project workspace instead of emitting an ENODEV extension stack after every failed provider attempt.
• Content/safety refusals are capped at one same-model retry when failover is unavailable, preventing the default transport retry budget from replaying the same refusal three times.
• Fable models remain visible in the model catalog, and a saved Fable default is honored when sticky-safety blocking is disabled.
• Claude models now omit discovered context files by default, avoiding provider false positives from unrelated instruction text; OMK_CLAUDE_CONTEXT_FILES=1 restores the full context.
• Extension resourceClaims now survive both tool-definition adapters, allowing dag-v2 to schedule non-conflicting custom tool calls concurrently instead of treating them as unclaimed exclusive work.
• Built-in tool-pair repair now uses the real AgentMessage contract (toolCall blocks and top-level role: "toolResult" messages), removing orphan pairs without unsafe message-shape casts.
• Pinned the transitive development dependency nanoid to 3.3.18, clearing GHSA-2v37-7h3g-55p8 from both full and production npm audits.

Docs

• Redesigned the root README around the Scope → Route → Verify → Replay control loop with a WCAG-aware cyberpunk OMK Girl hero and slow feature GIF generated through GPT Image 2. The root DESIGN.md now defines public brand tokens, media budgets, and reduced-motion guidance.

Removed

• Removed the grok-oauth-proxy provider path. Grok harness dispatch, failover, usage, and presets now use native xai. Stale models.json entries for grok-oauth-proxy are ignored instead of reappearing in /login and /model.

Release notes live in RELEASE_NOTES_v0.96.1.md.

Release v0.96.0

Added

• Added omk-protocol, the versioned TaskSpec -> ExecutionAttempt -> Observation -> EvaluationResult -> RuntimeDecision contract package, with runtime validators, explicit waivers, and pure semantic and runtime-decision reducers.
• Added evidenceReceiptToObservation() to project integrity-checked EvidenceReceipt v3 cores into immutable protocol facts. Legacy mutable EvidenceStatus and TaskContract verdict APIs remain compatible but are deprecated.
• Added the optional omk-book-to-skill package with compile/update commands, a pinned upstream workflow, advisory generated-skill scanning, and SHA-256 source/artifact provenance checks. Python extractors remain outside OMK core.
• Added deterministically seeded, bounded fast-check model and property suites for WPL transitions, replay migration and CAS, evidence freshness, subagent topology, run-journal CAS, and timeout/abort settlement ordering.

Changed

• New replay events declare jcs-rfc8785-v2 and hash RFC 8785-canonical payloads. Events without an algorithm remain verified as json-stringify-v1; mixed ledgers and exports preserve legacy records without rewriting them.

Fixed

• NVIDIA NIM's z-ai/glm-5.2 entry now transmits reasoning_effort, including the generated max thinking level; other NVIDIA models keep conservative compatibility defaults.
• Billing-cycle and quota exhaustion, including provider 403 usage-limit responses, now classify as provider.rate_limit and can switch to the first configured, authenticated resilience candidate before retry. Each attempt remains journaled, and a recovered retry ends with a later completed termination.
• Subagent DAG scheduling now sorts simultaneously ready lanes by lane ID, so topology aggregation does not depend on input insertion order.
• Local release bundles now include omk-adaptorch-wpl, allowing isolated installs of the full packed workspace without resolving that dependency from the registry.

Release notes live in RELEASE_NOTES_v0.96.0.md.

License

MIT