Skip to content

About ostk

Open-source project infrastructure for AI coding agents: durable working state, coordination, and trace that remain with the codebase.

PROJECT_CONTINUITY
Keep durable tasks, decisions, context pages, and audit records with the project so supported agents can reorient across sessions.
FLEET_COORDINATION
Coordinate ostk-managed agents through supervised lifecycles, durable project records, and explicit drain and recovery boundaries.
OPTIMISTIC_CONCURRENCY
Kernel-mediated CAS edits detect stale generations and surface conflicts. Native client and raw-shell writes remain outside that boundary.
CONTEXT_EFFICIENCY
Prompt caching and output compression can reduce repeated context work on supported routes; measured outcomes vary by model and harness.
CAPABILITY_BOUNDARIES
Configured pins and process sandboxes bound kernel-mediated work, with guarantees stated at the integration boundary.
IDENTITY_AND_TRUST
Project identity, trust tiers, and conditional signing make the provenance of governed and kernel-mediated work inspectable.

The Story

ostk wasn't designed in a planning meeting. It was extracted from the problem it solves.

The kernel started as a coordination layer for AI agents on the same codebase. First version handled file conflicts. Then process supervision. Then output compression. Then identity, heartbeat, audit trails, scheduling, drain snapshots, crash recovery, sub-stacks, and fleet-wide cache sharing.

The system grew through real multi-agent work. Each subsystem exists because a handoff, collision, recovery, context, or trust boundary failed in practice and needed a project-owned answer.

Built By

Scott Meyer.

I've been writing software a long time. The era doesn't matter — the principles do. ostk is built around local custody, explicit boundaries, durable state, and behavior you can inspect in the source and project artifacts.

Source Availability & Open Source Transition

The ostk source is published in public repositories on GitHub.

The runtime daemons and CLI orchestrators (ostk, ostk-cache, and ostk-recall) are open-source under the GNU Affero General Public License v3.0 (AGPL-3.0). This guarantees that modifications to these system elements remain available to the community.

The client IPC library (libostk) is released under the permissive MIT License. Developers can integrate the library directly into proprietary or closed IDE clients, pipelines, or automation loops without copyleft requirements. See License.

Real improvements get reported through the kernel itself. ostk should files them straight into the repo where they'll be seen.

Email works for everything else: [email protected]. I read everything.