Private development channel
2026-08-31
New ToggleLogic development remains an access-controlled Motherboard program until a separate ToggleLogic corporation is formed and public-release terms are approved.
Release policy
- Current owner. ToggleLogic is presently a division and product line of Motherboard, Inc.; ToggleLogic Inc. has not yet been formed.
- Private release approved. ToggleLogic Free 1.3.3 and ToggleLogic Intelligence 1.2.0-rc.10 are approved only for access-controlled testing and distribution by Motherboard.
- No new public package. These private versions are not being published to the public GitHub repositories, ClawHub, or npm. ToggleLogic Free 1.3.2 remains the latest generally available release.
- Future public release. Any later public release remains subject to corporate-formation planning and approval of the applicable license, contribution, patent, trademark, privacy, and distribution terms.
This notice changes the release channel for new development; it does not withdraw or modify the rights attached to previously released versions.
1.3.2
2026-08-28
A clearer adoption path, a sharper public boundary, and safer conversation continuity.
Changed
- Startup commercial use. ToggleLogic Free now ships under License 2.0, permitting qualifying startups to build, host, embed, distribute, and charge for commercial products while below the published funding and revenue thresholds.
- ClawHub introduction. The package now leads with configured routing, persistent owner overrides, local audit evidence, and local cost visibility, while stating the Free/Intelligence boundary more directly.
- Precise cheapest-default language. Free applies a deployment-declared static default; it does not classify requests or calculate a cheapest-capable model.
Fixed
- Conversation continuity. A short affirmative confirmation may inherit one preceding high-confidence routing decision only in the same session, within 15 minutes, and only once.
- Fallback isolation. Repeated model-resolution passes for one logical turn now pass through to OpenClaw so its fallback chain can advance normally.
- Fail-open behavior. An unmatched prompt leaves the host-selected model untouched, and required execution-surface metadata remains in the audit details.
Verification
- Full release gate passed. All 42 tests, syntax checks, release-identity checks, and package-boundary checks passed.
- ClawHub Plugin Inspector passed. Zero breakages were reported; the sole warning documents the intentional backward-compatible minimum host floor.
License 2.0
2026-08-28
Commercial startup adoption under objective funding and revenue thresholds.
Policy
- Build commercially while small. Qualifying startups may embed, host, and distribute ToggleLogic Free as an integrated part of a bona fide product and charge customers without advance contact.
- Objective thresholds. Contact Motherboard within 30 days after reaching either $1 million in cumulative outside funding actually received or $1 million in actual gross revenue during a trailing 12-month period.
- Operational continuity. A threshold event starts an automatic 180-day transition, followed by defined continuity for existing customers if no separate agreement is reached.
- Safer adoption terms. The new license removes at-will revocation, mandatory modification reporting, blanket patent exclusion for modifications, and immediate destruction before cure.
- Intelligence remains separate. The startup grant includes no rights to ToggleLogic Intelligence, the Toggle Registry, proprietary benchmark selection, private evidence, or other excluded components.
This is a license and documentation update. ToggleLogic Free runtime behavior is unchanged. The repository license controls.
1.3.1
2026-08-26
Current OpenClaw validation without abandoning existing supported deployments.
Changed
- Current build and validation host. The public package now records OpenClaw
2026.7.1-2 as the version used to package and validate ToggleLogic Free.
- Backward compatibility is explicit. ToggleLogic Free is backward-compatible with OpenClaw
2026.6.5 and later and is validated through OpenClaw 2026.7.1-2.
- The lower compatibility floors are intentional. Minimum gateway
>=2026.6.5 and plugin API >=2026.5.2 preserve existing supported installations; they do not mean ToggleLogic depends on an obsolete OpenClaw release.
Verification
- No runtime behavior changed. Routing, model selection, cost observation, fleet attribution, configuration, and network behavior are unchanged from 1.3.0.
- Release gate passed. All 40 tests, syntax checks, package-boundary checks, and the full quality gate passed on OpenClaw
2026.7.1-2.
- Artifact identity is public. GitHub and ClawHub publish the same 37-file artifact from source commit
313394a, SHA-256 6a27feb130aa0347a2959b8d0d89e06f1e8fe1f78c7789d7cc3d5f06c6665a81.
OpenClaw releases newer than the stated validation point should be verified before production promotion.
1.3.0
2026-08-26
Safer configured family routing, verified Intelligence compatibility, and profile-isolated operation.
Added
- Provider-constrained family aliases. Configured deployments can opt into model-family resolution using only explicitly approved providers already configured in OpenClaw.
- Verified Intelligence shadow mode. A separately licensed Intelligence installation can record recommendations without changing the model selected by OpenClaw.
- Reproducible release evidence. The release includes checksum, provenance, software bill of materials, and expanded security and isolation tests.
Fixed
- Uncertain pricing fails safely. Missing, stale, malformed, zero, or incomplete pricing can no longer produce a family selection or a misleading zero-dollar estimate.
- OpenClaw profiles remain isolated. Named and development profiles keep ToggleLogic logs, pricing caches, and state inside the active profile.
Safety
- Explicit choices remain authoritative. Owner and user session model selections stay above automatic routing.
- The Free/private boundary remains intact. The public package contains no private classifier, proprietary model registry, customer evidence, credential, or deployment backup.
Verified with 39 passing Free tests, 105 paired Intelligence tests, independent review, a Headquarters shadow canary, and a clean ClawHub Plugin Inspector report.
1.2.4
2026-08-23
Shadow-mode routing audit support. Host routing remains unchanged.
Added
- Family-resolution audit detail in shadow mode. When the separately licensed Intelligence layer is evaluated in shadow mode, ToggleLogic Free records its structured model-family recommendation in the local audit record without returning a model override to OpenClaw.
Safety
- The public and private boundary remains intact. The Free package contains no private classifier, benchmark data, prompt content, customer data, provider credential, or model registry.
- Shadow mode does not change routing. It is an observation and audit gate only; the host keeps control of the selected model.
Verified with 21 passing tests and a clean ClawHub Plugin Inspector report. Release source: commit 88fcd317.
1.2.1
2026-08-22
Release-provenance correction. Runtime behavior is unchanged from 1.2.0.
Fixed
- Release provenance now points to the exact source commit. The CycloneDX software bill of materials now links to the exact public GitHub commit for this release. Version 1.2.0's plugin archive remains intact; this patch corrects its immutable SBOM metadata only.
The source, package, OpenClaw manifest, runtime identity, checksum, and SBOM now identify one exact release commit.
1.2.0
2026-08-22
Public routing-contract hardening. The Free plugin remains deliberately separate from the licensed Intelligence product.
Fixed
- Configured static routes now honor structured task labels. In configured mode, routes can match a host-supplied task label and then an optional
default. The Free plugin does not inspect or classify message text to infer a task.
- An explicit Intelligence selection is no longer silently downgraded. When the separately licensed Intelligence layer is unavailable or incompatible, ToggleLogic records that state and passes through to the host instead of quietly switching to the cheap-default mode. Automatic mode retains its documented safe fallback behavior.
Safety
- Public and private responsibilities remain separate. This release contains no private classifier, model registry, learning system, customer evidence, provider credential, or private Intelligence source.
- Release identity is tested. Package metadata, the OpenClaw manifest, and the runtime version must agree before release validation passes.
The test suite passes 20 of 20 tests, and the ClawHub Plugin Inspector reports zero issues and zero warnings.
1.1.2
2026-08-21
A fail-closed compatibility contract for the separately licensed ToggleLogic Intelligence layer. Free routing behavior is unchanged.
Fixed
- Runtime and package identity now agree. The version reported by the running plugin is aligned with the package and OpenClaw manifest version.
Safety
- Private Intelligence compatibility fails closed. The Free plugin accepts a separately licensed Intelligence installation only when its release manifest is released, version-consistent, declares seam ABI 1, and explicitly includes the running Free-plugin version in its compatibility interval.
- The public/private product boundary remains explicit. This release contains no classifier, model registry, learning system, customer evidence, or private Intelligence source. When Intelligence is absent or incompatible, the Free plugin continues without activating it.
Existing Free installations require no configuration change. The test suite passes 14 of 14 tests, and the ClawHub Plugin Inspector reports zero issues and zero warnings.
1.1.1
2026-08-14
Current ClawHub manifest compatibility. Routing and runtime behavior are unchanged.
Fixed
- Removed unsupported manifest metadata. The package no longer places
license and categories in the OpenClaw plugin manifest. License information remains in the supported package metadata and package documentation.
- Validated against current OpenClaw targets. ClawHub package validation passed with zero issues and warnings on OpenClaw
2026.8.1-beta.1 and 2026.7.1-2.
No configuration or behavior change is required for an existing 1.1.0 installation.
1.1.0
2026-08-11
Privacy-safe local fleet metering. Routing is unchanged.
Added
- Deployment attribution. Local cost-log rows can carry a configured non-secret deployment identifier and optional cost-center slug. If omitted, the local hostname is used.
- A versioned fleet-ledger contract. New rows identify their compatible local usage schema, enabling a headquarters collector to validate records without reading prompt or assistant content.
- Invoice-safety markers. Local public-rate estimates are explicitly marked as estimates and not invoice-ready. Provider reconciliation is still required before a customer can be billed.
Safety
- Local only; no phone-home telemetry. ToggleLogic does not send usage, deployment identifiers, cost centers, or customer information to Motherboard or any ToggleLogic endpoint.
1.0.6
2026-08-02
A priced call with no usage was reporting $0.00. It no longer does. Routing unchanged.
Fixed
- A priced call with no usage no longer reports a false $0.00. The cost observer computed a dollar figure any time the model was priced, without checking whether the call actually reported token usage. When a priced call came back with no usage, or with non-finite usage, the missing tokens were treated as zero and the result was recorded as
costUsd: 0.00 — a reported number, indistinguishable from a call that genuinely cost nothing. This shipped with cost visibility in 1.0.3 and has been present in every release since, including the 1.0.5 available until today.
- Three distinguishable outcomes instead of two. A priced call with valid usage records its real cost. A priced call with missing or non-finite usage goes loud — marked
usageMissing, reason priced-but-usage-missing, with no dollar figure recorded. An unpriced model stays loud, exactly as before.
- The daily tally counts it correctly. Missing-usage calls now land in the loud bucket rather than in priced spend, so a day full of them can no longer read as $0.00.
Changed
- The named regression test now asserts the complete behavior. The existing test checked only the missing-price half; the missing-usage half that produced the false zero had never been tested. A new regression test covers all four cases — priced with usage, priced without usage, priced with non-finite usage, and unpriced. It fails on 1.0.5 and passes on 1.0.6.
Compatibility unchanged: minimum OpenClaw >=2026.6.5, validated through 2026.6.11. OpenClaw's plugin manifest carries a minimum floor only — there is no ceiling field — so the validated upper bound is documented here rather than enforced by the manifest.
1.0.5
2026-07-15
Owner-override expiry — a pinned model can now expire on its own, and a malformed expiry fails safe. Routing otherwise unchanged.
Added
- Optional owner-override expiry (
expires_at_ms). An owner-override may carry a numeric epoch-millisecond deadline. Once it passes, the override stops applying and routing returns to automatic selection — no file rewrite required. Overrides with no expiry behave exactly as before (they hold until cleared), so this is fully backward-compatible.
Fixed
- Expired overrides are no longer honored. An override past its
expires_at_ms deadline is now rejected; previously it was still applied.
- A malformed expiry fails closed. If
expires_at_ms is present but not a finite number, the override is rejected — never silently honored — and the rejection is recorded on the structured audit stream.
- Plugin version string corrected. The reported plugin version was stale; it now matches the package version.
1.0.4
2026-07-09
Packaging metadata fix — routing and cost visibility unchanged.
Fixed
- Corrected the plugin manifest description. A packaging-metadata field had been truncated; it now carries the full, accurate plugin description. No change to routing, pins, or cost-visibility behavior — identical to 1.0.3 in operation.
1.0.3
2026-07-06
Cost visibility — your calls priced live in dollars, and a missing price can never quietly read as $0.00.
Added
- Live dollar cost, per model and per day. A new opt-in, observe-only capability reports what your calls cost in dollars, priced from dynamic public rates fetched live from Models.dev — cached locally and refreshed on a slow cadence, never fetched per call. Curated to the major providers' standard lineups (Anthropic, OpenAI, Google, xAI, Meta).
- Bundled offline fallback. If the live pricing source is unreachable, pricing degrades to a bundled offline snapshot (LiteLLM, MIT) — last-known-good, never a crash and never a silent $0.00.
Changed
- Unpriced usage is loud, never a silent $0.00. Any model the curated free-tier set can't price is reported explicitly as unpriced — with model name and token count, shown separately — and is never rolled into the tracked dollar total as $0.00. Priced and unpriced always appear together.
Safety
- Observe-only — cost visibility reports, never blocks. It never halts, blocks, or downgrades a request. Spend enforcement is a paid-tier capability, so a pricing gap can never interrupt your calls.
1.0.2
2026-07-05
Publishing + discoverability corrections.
Changed
- Package now published under the ToggleLogic organization on both ClawHub and GitHub, with the source repository correctly linked to the ToggleLogic org.
- Release notes and compatibility metadata aligned across the ClawHub listing and the GitHub release.
Fixed
- Corrected the source-repository link that previously pointed at a personal account, resolving a listing/recognition issue.
1.0.1
2026-06-23
Pin reliability + packaging cleanup.
Fixed
- Owner and session model pins now hold reliably above routing — a pinned model takes precedence over automatic route selection.
Changed
- Packaging metadata cleanup.
1.0.0
2026-06-18
Initial public free-tier release.
Added
- Free-tier model routing for OpenClaw: owner overrides (sticky, held), static/intent routes, and a deliberately simple cheapest-default. Bring your own provider credentials, declare intent, and let the plugin apply your choices.