Anatomy of a
super agent
TeamX doesn’t replace the agents you love — it fuses them. Local CLIs, remote runners, and raw API models become limbs of one organism with one memory, one scheduler, and one mission: ship your product.
Engines in. Software out.
Connect every engine you have and TeamX becomes the nervous system between them — six organs turning raw agent hours into shipped product.
in · your engines
Local agents
Remote agents
API · BYOK
the brain · six organs
each job → the engine that fits
work slotted into compute windows
every run budgeted, never wasted
one memory shared by every engine
every change written down first
tests + blast radius before merge
out · shipped software
Smart router
each job → the engine that fits
Every task is profiled before it runs — token size, surface sensitivity, capability needs, urgency — and matched to the best engine that is awake and on budget.
›TMX-318 · ~41k tokens · auth surface → claude code (local)
Epic scheduler
work slotted into compute windows
Your roadmap stops being a list and becomes a schedule. Epics are decomposed, dependency-ordered, and slotted into the hours where compute actually exists.
›04:00 window reset → TMX-101 heavy refactor un-paused
Token governor
every run budgeted, never wasted
Every run gets a budget before it starts and a bill when it ends. When a window runs low, new heavy work queues instead of starting on fumes.
›claude max at 71% → heavy tasks queued · light tasks continue
Radial context
one memory shared by every engine
The agent picking up task #4 already knows what happened in tasks #1–3, which decisions were made, and why. No cold starts, no re-explaining your codebase.
›begin_task → affected files · recent commits · prior decisions
Spec layer
every change written down first
Nothing touches code before a written spec: problem, approach, blast radius, success criteria. You review intentions, not diffs.
›begin_task refuses unspecced work — bypasses are logged
QA gates
tests + blast radius before merge
Every task closes with a test report. Sensitive surfaces demand a human gate that blocks the epic until it passes.
›epic TMX-92 held: QA gate red on /billing surface
Watch it think
Every task is profiled for size, surface, urgency, and budget — then handed to the engine that fits, never the engine that's merely next.
Every task profiled first
Token size, surface sensitivity, capability needs, urgency — measured before a single token burns, and turned into a routing rule you can read.
incoming task · queue live
Routed to the engine that fits
The gnarly auth refactor goes to Claude Code on a fresh window. The docs pass runs free on your Ollama box. Nothing defaults to whatever's next in line.
engines · routing target
Claude Code
local cli · flagship
Codex
local cli
Claude → GitHub
remote
Anthropic · OpenAI · OpenRouter
api · byok
Ollama
local api · free
five real routing decisions
Assembled in two minutes
No SDK, no repo changes, no YAML weekend. Your agents already speak MCP — TeamX is just the brain they plug into.
Add the MCP server
One command in your agent’s config. Claude Code connects today; Codex and friends speak the same protocol. Nothing installs in your repo.
Authorize once
Your agent signs in with a scoped token and appears in mission control with a live status dot. Revoke any agent’s access in one tap, any time.
Approve your first spec
Drop an idea. TeamX interviews it into a spec; you approve it; the deck lights up. The loop is running — and it doesn’t stop when you do.
Your keys, your subscriptions, zero markup. TeamX never resells tokens — it can’t profit from your burn, which is exactly why the governor works so hard to minimize it.
Your agents are idle right now
Every hour your subscriptions sit unused is shipped software you didn’t get. Wake them up.