Code Command Center shipped hard this cycle — from a distributed compute fleet to a council of rival minds to agents that evolve their own strategy. Here's what left the launchpad.
Each one ships behind a flag, wired into the ledger, and safe to run against a live portfolio. Together they turn CCC from a session dashboard into an orchestration substrate.
GPU/VRAM-aware distributed agentic compute. Enroll machines, dispatch agent work across the fleet, and route each task to the node whose loaded model is strongest for the job — with hard org/project isolation and a fresh dispatch-time readiness probe on every hop. Ships with a web console and a standalone node binary.
CCC was raised by a single opinionated mind. Now it keeps a Rolodex of them — Claude, OpenAI, Gemini, Groq, and Grok (xAI), plus your own local models. Route a task to the strongest mind for it, or convene a panel to argue the hard calls before you ship. Cost ledger and offline-clamp built in from day one.
CCC that improves CCC. Agent skill-strategies compete champion-versus-challenger across golden eval sets; tournament judging fans out over the worker fleet; and only a gated winner earns promotion — with last-known-good rollback the moment a champion regresses. Optional, off by default, strongly gated.
A forward-looking "what to work on next" surface. Sonnet writes a structured record as every session ends; Opus organizes the whole org each morning into a feed of trajectories, risks, and suggestions — now with a Yesterday recap and title-stable dedup so themes update in place.
Visual DAG orchestration of agent sessions. Watch any node's live terminal inside the run viewer, replay historical runs, and click a connector to inspect it — now on mobile, with per-pipeline Anthropic credentials and run/deep-audit observability.
Ask your repo "do we already do X?" in plain English and get a cited answer with a clickable file:line — one it re-reads at answer time and shreds if the code isn't actually there. It would rather say "I couldn't find that" than invent a function that doesn't exist.
An entertaining, multi-host audio briefing of your entire portfolio. A cast — anchor, builder, skeptic, comedian — trades the week's headlines over real voices you can clone, on a schedule, capped by a monthly state-of-the-union.
The newest Claude models land in the registry — including a Fablonnet plan mode: Fable drafts the plan, Sonnet 5 executes it. Pick the right mind for thinking and the right one for doing, in the same session.
A Linear-style launcher wired straight into CCC. Rename or kill the focused session, add a repo, push a knowledge entry, flip a setting — every action a single keystroke away.
Screens pulled straight from this cycle's pull requests — captured against a live, isolated instance.
Every pipeline run renders as a live DAG — gates, loops, and branches lighting green as they pass. Drop into any node's terminal without leaving the view, then replay the whole thing from the timeline once it lands.
Behind Codebase RAG sits a full structural map — thousands of nodes and edges, LLM-refined, regenerated on demand. It's what lets an agent answer "do we already do X?" with a real, clickable citation instead of a guess.


One operator now conducts a fleet of machines, a council of models, and a swarm of agents that tune their own playbooks overnight — while the ledger counts every token and the citations reconcile themselves.
Nine more capabilities that landed quietly this cycle.
Launch computer-use agents as first-class targets, now with a native Windows display backend.
Run the smart-glasses bridge on a private, local, or remote model via env override.
Every session mints a hash-only token for headless user attribution, with revoke + expiry sweep.
Pin project & repo checkouts to their branch so nothing drifts off main mid-run.
Boot a read-only standby instance alongside prod for zero-write observation.
Per-automation ownership with a full claim history — no more anonymous cron jobs.
Split the Live sessions panel by who — or what — started each session.
First-pass-success and retry-burn economics, straight off the passive usage ledger.
Promote a one-off session learning into a durable, project-wide rule in a click.
Behind the headline launches, the cycle poured itself into durability: dozens of refactors folding duplicated logic into single sources of truth, a wave of mobile-UX and icon-consistency fixes across every page, and new test suites guarding the load-bearing paths. Power you can trust is power that keeps working on the tenth run.