Sidecat Not built. Named so it cannot be quietly forgotten, not so it can be claimed.
Roadmap
The road to operations agents can trust.
Sidecat is early, but the center is settled: agents need a workbench where they can start from a record, run tools as structured operations, remember the useful path afterward, and ask what evidence makes the current context safe to trust.
The native line is what runs today: stable operation IDs, typed payloads, sidecatd routing, governed commit and push attempts, file-backed Lisp workbench routines, API request and OpenAPI discovery support, and a review-to-planning loop split across Sidebird and Sidebar. Sideinfra adds the source-side intelligence: accepted versus merely present, proven versus only claimed, current versus stale.
The standard
Demonstrated, not asserted.
A capability is not counted as working until it has been run end-to-end on real, messy work. A green unit test, a wrapper, or the mere existence of a command does not earn the claim — only the real thing, exercised the way it is actually used.
Today the proof is a local loop around a real external effect: a browser-authenticated operator, a credential grant held by sidecatd, an AAuth interaction, and a manually run GitLab push to sidecat-lorem. The next proof is broader: the same idea should work through an operation catalog, a small workbench language, Sideinfra trust-gap checks, and hosted sidecat.org feedback.
The maturity ladder
Six rungs, each one earned from the tree.
Every capability is placed on one scale and moved up it only by what is actually built. The current Sidecat line is still below a public-use bar; the point of the next milestones is to make one useful operation path real enough that agents and humans can build on it without pretending.
New and just barely usable — enough shape to grow, and enough to live on while it does. An experimental floor, never a release bar.
Works end-to-end for its primary purpose, on real data, exercised the messy way it is actually used. Refined errors, scale, and real-user testing still lie ahead — those three are the next bar.
Feels like a capability, not a crutch. It works, with a lot of room to improve.
Everything works: extremely reliable, extremely capable.
Goes above and beyond — a genuine change in how agents and people do work together.
The standard is practical: can an agent understand what exists, run it safely, inspect what happened, and save the useful path for a later agent?
The milestones
From local proof to reusable operations.
The order is current foresight within the real dependencies, not a fixed schedule. Each milestone should make the workbench and intelligence layer more useful without expanding the command tree.
The native line makes operations the source of record: stable IDs, typed contracts, effect metadata, authority needs, runtime location, evidence, and receipts.
The workbench line makes useful paths reusable: Sidecat Lisp sessions, scripts and macros, package promotion, sidechat coordination, and Sideinfra checks that help agents change the codebase without drifting.
Sideinfra reads sidecatd projections, repo facts, and profile declarations, then derives calibrated gaps such as accepted-but-unverified, present-but-unowned, or proven-but-stale.
The first stable public line should let agents start work, run structured operations, remember what happened, and reuse the path across local and hosted Sidecat surfaces.
Already standing
What is already standing.
These are the public claims the project can point at today. They are still early and source-first, but each one has repo-owned code, fixtures, tests, public writing, or live proof behind it.
Start, run, remember
The public product direction is a workbench where agents can keep context, run tools, and save what happened for review and reuse.
File-backed workbench routines
Project .lc files can feed the Lisp workbench with source digests, call sidecatd operations, and become the durable starting point for package authoring instead of living as pasted commands.
Local AAuth operator proof
The local Person Server authenticates the operator, handles browser approval, and lets sidecatd spend an admitted GitLab credential after the interaction is approved.
Operation catalog pivot
Operations are now the stable unit underneath the CLI, packages, daemon calls, SDKs, web surfaces, and Sidecat Lisp.
Hosted Sidecat planning
sidecat.org is being shaped as the hosted account, approval, credential, replay, messaging, and federation surface for real daily feedback.
Sideinfra first slice
The first Sideinfra explain path now derives work-subject judgments from sidecatd projections and reports the remaining proof or currency gap before an agent acts.
Review-to-planning surfaces
Sidebird reviews what happened; Sidebar helps plan and approve what should happen next. A clarified Sidebird review is promoted into Sidebar planning context as a handoff, not a milestone.
This page records the fourth-generation plan. The repository's main line is now the Gen5 clean-sheet implementation: its accepted architecture and current maturity live in the Gen5 architecture maps, and its release boundaries live in BUILD.md. The architecture page walks the fourth-generation spine those milestones built.