Every team searches for the same capability again.
People hunt through repositories, compare near-duplicates and repeat installation work for every agent environment.
Unlimited Skills is an independent, local-first control layer for discovering, governing, routing and distributing capabilities across a mixed fleet of AI agents.
Your company may need 500 skills. Each agent needs only the right one now.
Reproducible project measurements, not universal performance promises.
A useful skill is easy to install once. It becomes an operational system when the company has dozens of roles, hundreds of agents and several providers.
People hunt through repositories, compare near-duplicates and repeat installation work for every agent environment.
Unsafe instructions, prompt-injection patterns, dependencies and updates must be checked repeatedly. An approved version can quietly drift from the version agents actually run.
Loading every skill and every tool schema into standing context spends attention and tokens before the task has even started.
Manual copies do not answer which agents received a pack, which version they use, whether policy was applied, or where an update failed.
The old workflowFind it. Install it. Review it. Copy it. Review it again. Hope every agent has the same version.
Unlimited Skills separates the company’s capability library from every individual agent context and every individual vendor environment.
Bring together local skills, approved repositories, public collections, private packs and MCP tool definitions without moving agent workloads into another platform.
Keep approved, blocked, local-only and needs-review capabilities separate. Allowlist-only distribution prevents the hub from silently falling back to an unaudited catalog.
Define which collections or private packs belong to which organisation, team, installation, agent type and release channel.
The agent searches at each substantive work phase, receives at most a few candidates and loads the one capability needed now instead of carrying the whole library.
Per-pack acknowledgements, retry state and compliance receipts will prove which version was applied to each managed agent, not merely that the installation was recently online.
Sales may use one provider. Engineering another. Operations may run open-source and custom agents. Unlimited Skills is designed as a layer above those runtimes, not as another runtime you must migrate into.
Provider coverage expands through adapters. The independent library and policy model stay the same.
The open-source router solves capability overload today. Business and Enterprise editions turn that foundation into a governed delivery system.
For individuals and small teams that want the full local routing model without registration.
For small teams learning how governed capability sharing works before they need private distribution.
For AI-native companies that need one operating source of skills across departments, runtimes and release cycles.
For large mixed-agent fleets where capability delivery is part of security, compliance and change management.
Status is explicit by design: the MIT core works today; hosted collaboration, customer-facing fleet operations and some Business/Enterprise controls are alpha or in development. We do not present roadmap features as finished production software.
The goal is a reusable approval decision: one reviewed capability, one governed version and an evidence trail showing where it is allowed to go.
Unlimited Skills does not claim that every skill is safe. It creates the controls needed to keep unreviewed, unsigned or policy-incompatible material away from managed distribution.
Community search, routing and the library stay on the user’s machine. Local prompt text is not sent to a hosted service by the core.
Approved, blocked, local-only and needs-review capabilities remain distinguishable instead of collapsing into one trusted-looking catalog.
Registered paths verify signed manifests, SHA256-pinned downloads and path-safe archive extraction before installation.
Enterprise Skill Lock can constrain registries, channels, signing keys, local roots, community actions and remote fallback.
A catalog helps people find skills. A control layer keeps capability choice, delivery and policy coherent after the company scales.
No. It sits above supported agent runtimes as an independent capability layer. Your teams keep the providers, models and custom systems that fit each job.
Yes. The local Community Core is MIT-licensed, offline-first and does not require registration for local search, routing, indexing or use.
No. Local search and routing stay local. Hosted actions are explicit, registration-gated and bounded by the documented contract.
The technical foundation exists across team sync, signed manifests, private-pack clients, policy controls and governance metadata. The complete customer-facing Business and Enterprise product, including delivery receipts, encryption and the fleet dashboard, is still being productised.
Yes. We are opening a design-partner track for organisations with mixed agent fleets, private skills and real governance requirements.
Tell us where skills live, which agent runtimes use them and what your approval boundary looks like. We will map the shortest credible Business or Enterprise pilot.