Skip to content

License keys and limits

Referenced by Section 3.2 of LICENSE.md. This page describes the tiers and limits that apply to the current version; the Licensor may set different limits for future versions and for the keys it issues. Limits are delivered to a deployment by the ChannelGate platform, so they can change without a new release; the limits that apply to an enterprise key are fixed for its agreed term.

Tier How you get it Registered channels Registered users Active chat threads per channel per UTC month AI runs per channel per UTC month
No key install and run 10 (one active channel per month) 10 100 in the active channel 500
Registered (free key) create an account with your email on the ChannelGate platform 10 10 100 in each channel 500
Enterprise key Enterprise License with Makeitfuture unlimited by default unlimited by default unlimited by default unlimited by default
  • A channel is a persistent channel, group or direct-message record on a supported chat platform. A conversation for the 100 allowance is an active chat thread within that channel, not another channel and not an individual AI message. Each channel has its own UTC-month allowance. Repeated messages and automation in the same admitted thread reuse its slot.
  • Registered channels include DMs, groups and the synthetic API conversation. Registered users are all stored human account records, including unapproved users; the synthetic API principal is excluded. Inventory counts do not reset monthly. Non-Enterprise deployments cannot create an 11th record of either kind. Existing records remain editable and usable after an upgrade or downgrade; growth is refused until below the cap, and other limits still apply.
  • Free installations allow 100 active threads per channel per UTC month, 10 channels and 10 users. Legacy or newly signed Free payloads cannot raise these compiled ceilings by omitting fields, supplying null, or granting larger values. Signed smaller caps are honored. Enterprise defaults to unlimited and may carry agreed limits.
  • Without a key, the first active channel admitted in a UTC month remains the one served that month; its thread allowance is 100. A registered Free key allows activity in its registered channels, each with its own 100-thread allowance.
  • Threads already admitted before a stricter limit or Enterprise downgrade remain usable for that UTC month. New threads are refused once their channel’s allowance is reached. The set resets at the next UTC month boundary; admitted threads still obey the AI-run allowance.
  • An AI run is one engine run started for a channel. User turns, follow-ups, schedules and background jobs share the existing 500-run allowance per channel per UTC month. Memory-review runs are excluded. The 100-thread allowance does not grant 500 runs to each thread.
  • Thread tracking begins with post-upgrade activity. The upgrade does not fabricate past thread history: an old thread counts when it is next used, and existing channel AI-run counts remain.
  • At a monthly limit the assistant answers with a short notice instead of starting a run. The existing AI-run allowance warns once at 80 %.

Standalone automation fires without an existing chat thread use a stable logical conversation identity for that fire. Once a delivery thread is known, later runs in that thread share its slot. Background agents and question continuations in an existing conversation keep the parent conversation identity even when their engine sessions are isolated.

A key belongs to the organization that operates the deployment — an end-user key. An agency or service provider that installs or operates ChannelGate for a customer under Section 3.1 uses the customer’s key: the customer creates the account with its own email and may add the provider as a manager. Keys are never shared, pooled, or transferred across organizations.

Key verification runs at start-up and about once a day; usage is reported as counts. The payload for verification contains the raw license key, installation id and software version. Usage reports contain a key hash and per-conversation hashed ids with message counts. Licensing payloads never contain message content, user identities, channel names, provider credentials or files. The stable hashes are pseudonymous, not guaranteed anonymous. The full data-flow description is in PRIVACY-AND-DATA-FLOW.md.

A deployment that cannot reach the platform keeps its last verified tier for 14 days and shows a banner in the admin UI. After the grace period it falls back to the no-key limits at the start of the next calendar month — never mid-month, never silently. An unlicensed or unreachable state is a healthy state: the daemon keeps running and no run is ever killed by the license check.

The Licensor may raise or lower the no-key and free-key limits for future versions and for keys it issues after the change; changes are announced in CHANGELOG.md.

License verification and enforcement live in src/ee/. That directory is proprietary, source-visible code owned by MAKEITFUTURE S.R.L. — it is deliberately readable so that an operator can audit exactly what the check does and what leaves the install, but it is not covered by the Sustainable Use License in LICENSE.md. Its own terms are in src/ee/LICENSE-EE.md, and Sections 3.2 and 4.5 of LICENSE.md make removing, disabling, or circumventing what it implements a licence violation.

File What it is
src/ee/LICENSE-EE.md the terms for the directory
src/ee/tiers.js the platform URL, the compiled-in no-key limits, the canonical JSON the signature covers
src/ee/license-public-key.js the Ed25519 public key every license payload is verified against
src/ee/license.js key storage, installation id, verification, the state machine
src/ee/limits.js conversation admission, the monthly cap, the usage report

Inventory admission is checked atomically inside the user/channel store transactions before new records are written. licenseAdmission() is called by the run orchestrator (src/gateway/run.js) before a turn provisions a folder, mints a session, or spawns an engine. A refused turn is a short reply, never an error and never silence.

Any one of these; the first that is set wins.

  1. Admin UI — Settings → License → License key, then Verify now. The value is stored in settings.json, never returned by a listing (only hasLicenseKey and the last four characters), and can be read back one at a time through the same password-protected reveal as every other secret.
  2. Chat — an admin can use the set_license_key gateway tool (and clear_license_key). Both need a human Approve click; get_license_status is readable by anyone allowed in the channel and never shows the key.
  3. Environment — CHANNELGATE_LICENSE_KEY, the bootstrap source for a container that has never had an admin session.
Variable Meaning
CHANNELGATE_LICENSE_KEY the key, when it is not set in the admin UI
CHANNELGATE_PLATFORM_URL the platform base URL (default https://channelgate.dev); also settable in the UI
CHANNELGATE_LICENSE_PAYLOAD a signed offline license (see below)
CHANNELGATE_LICENSE_PUBLIC_KEY overrides the compiled-in production verification key (staging or an explicitly coordinated key rotation only)

A deployment with no route to the platform can be issued the signed payload directly and set it as CHANNELGATE_LICENSE_PAYLOAD (the JSON {"license": …, "signature": …} the verification endpoint would have returned, raw or base64url). It is verified locally against the same public key, so it can only ever say what the Licensor signed and it stops working at its own expiresAt. An install running on one makes no licensing request at all — no verification and no usage report.

The gate is a small state machine. Every state is a healthy daemon state: no run is ever killed, crashed, or silently dropped by the license check, and every state other than no_key and valid raises a banner on the admin License card that says what happened and what to do about it.

State When Limits in force
no_key no key configured no-key tier
valid the last check succeeded and the payload is in date the payload’s
invalid the platform returned 401 invalid_key no-key tier, immediately
revoked the platform returned 403 revoked no-key tier, immediately
expired the licence passed its own expiresAt no-key tier, immediately
grace the platform is unreachable, within 14 days of the last success the last verified tier (no-key tier if there has never been one)
expired_grace unreachable for more than 14 days the last verified tier until the next UTC month boundary, then the no-key tier

invalid and revoked drop immediately because the platform has positively stated that the key is not valid — that is a Section 3.2 statement, not a network problem. expired drops immediately for the mirror-image reason: a licence that ran out on its own terms did not lose contact with anything, and its end date was known in advance, so the offline courtesies do not apply to it. This is checked before the unreachable lane and holds for an offline payload too — an air-gapped deployment’s payload stops granting its tier the moment it expires, exactly as this page has always said it would. Only the unreachable path gets the month-boundary courtesy, because that failure is usually the Licensor’s, not yours. A response whose signature does not verify is treated as unreachable in both directions: it can neither raise nor lower a tier.

  • The existing (UTC month, channel id) license_usage ledger counts AI runs and legacy active-channel admission. Its conversations signed limit retains this channel-ID meaning for compatibility; it is not the 100-thread product allowance. The Free default of 100 there is redundant with the 10-channel inventory cap; stricter signed active-channel limits still apply.
  • conversationsPerChannelPerMonth is the signed thread allowance: 100 for Free and no-key, unlimited (null) by default for Enterprise. The separate monthly thread ledger is keyed by channel and the gateway’s resolved thread identity. Platform-qualified channel IDs prevent collisions between identically named channels on different platforms.
  • Admission and counting happen before engine spawn, atomically, so a refused turn does not count and concurrent first messages cannot exceed a cap. A run without a thread identity uses the gateway’s stable unthreaded identity in that channel; it cannot create unlimited slots.
  • Memory-review runs are excluded, by origin (memory_review) and by toolset (CG_TOOLSET=memory-review).
  • The 80% AI-run warning is posted once per channel per month, prefixed to the answer the user is already receiving — it costs no extra message and cannot be missed.
  • Admin UI: Settings → License — tier, key last four, when it was last verified, when the next check is due, the state banner, and this month’s per-channel AI-run and active-thread usage with their limit lines.
  • Chat: the get_license_status gateway tool.
  • API: GET /api/license (admin session), and POST /api/license/verify to check now.

Enterprise integrations and support status

Section titled “Enterprise integrations and support status”

Composio SDK mode is Enterprise-only and Beta, enforced when settings, sessions and bridge requests are processed. The standard Composio MCP integration remains available in every tier. Microsoft Teams and Google Chat connectors are Beta, with platform-specific live acceptance required before relying on them for unattended work. Tier changes never weaken core security.

Organization admins can configure Approved harnesses in the admin channel editor or Slack Settings → Agent. Leave the policy unrestricted, or approve one or more known harnesses. The channel default must be approved. Channel selection, explicit thread pins, saved sessions, per-run overrides, queued work and automatic failover are checked against the saved policy. A disallowed selected harness produces a notice; it is never silently substituted to bypass a pin. Background/scheduled/API turns use the same launch boundary. The Claude memory reviewer is skipped when Claude is not approved.

Only an effective Enterprise license permits changing the policy. A saved restriction remains enforced after downgrade or expiry; restoring Enterprise is required to change or clear it. The default on existing channels is unrestricted. This governs gateway-managed harness launches, not arbitrary commands a Worker/Admin author is otherwise allowed to execute inside a container.

Non-Enterprise installations expose one month (30 days) of cost history in audit and dashboard reports. Enterprise history is unlimited. Older accounting remains stored for upgrades; license usage counters and conversation history are independent. Structured costs older than the reporting window are also hidden in audit-event, API-job and schedule-telemetry responses, while their operational records and result text remain available.

The following capabilities require an effective Enterprise license (including use of previously saved configuration after downgrade or expiry):

  • Skill preferred-model switching: opt a conversation into honoring a skill’s engine/model preference.
  • Exact company email-domain auto-approval for new users and policy application to existing users.
  • A dedicated Codex login per conversation, separate from the gateway login.
  • The external skill-catalog MCP endpoint and its catalog/user access tokens.
  • Organization environment variables shared across conversations.
  • Google Drive workspace synchronization, both scheduled and on demand.
  • Channel skill templates and their inherited skill grants.
  • Per-channel monthly USD spending limits.

Other tiers retain ordinary skills, channel/personal environment variables, the shared gateway Codex login, manual user approval and standard chat connectors. Stored Enterprise configuration is preserved when an entitlement expires; it does not activate Enterprise behavior. A channel configured with a dedicated Codex login is refused rather than silently switched to another account.

A spending limit is an admission control against the channel’s recorded spend in the current UTC calendar month. It is not a provider billing limit: an in-flight run can take recorded spend above the cap, concurrent runs can overshoot, and provider-side costs are not reserved in advance. monthlySpendingLimitUSD: null means no cap; a configured cap is a positive finite USD amount. A current-month usage row with unknown cost refuses new runs until pricing is repaired or an admin clears the cap. A saved cap also refuses new runs after downgrade; an admin can clear it without Enterprise, so the safety control never silently disappears. See the regression checks under Enterprise organization controls in TEST-PLAN.md.