outsourcerer

---

You already pay for a fleet of AIs: Claude, Codex, Gemini, maybe Devin, a stack of OpenRouter credits. Each is brilliant at something the others aren't, and they sit in separate rooms. Outsourcerer makes them work as one team. It sends the boring work to the cheapest engine that can nail it, pulls a top-tier model in when the job actually needs a big brain, keeps a running receipt of what that saved you, and, uniquely, clones your setup onto whichever engine runs the job.

It doesn't just call another model. It brings your whole workshop.

When Outsourcerer hands a job to a different engine or harness, that engine doesn't show up empty-handed. It carries your setup with it: the same skills, the same plugins, the same MCP servers your main agent uses. A cheap model running on Devin can use your custom skill and reach your MCP tools exactly the way your Claude session would. It's the difference between borrowing a stranger's bare laptop and having your own, fully set up, wherever the work happens to run.

---

The opportunity hiding in your subscriptions

You're already smart about this. Even inside Claude Code you hand the small stuff to Sonnet instead of Opus. Good instinct, Outsourcerer just takes it further than any single harness can:

• What if that same errand could run on an even cheaper engine, sometimes near-free on a subscription you already pay for, at quality that rivals your frontier model?
• What if a couple of stronger models could sanity-check your plan before you build, in parallel, without you leaving your session or wiring up a thing?

The models and the subscriptions are already on your machine. You could stand up a full multi-agent system to connect them, and sometimes that's exactly the right call. Outsourcerer is the lighter path for when you just want the work moved to the right engine and the savings counted.

---

You talk. The sorcerer handles it.

No flags to memorize, no routing tables. You say what you want; it checks what you actually have installed, picks the right engine, and offers you the cheap path before spending a cent it didn't have to.

If something isn't installed, it tells you the one command to fix it. You just say yes.

---

Advisors: bring the big brains in, several at once

Delegation runs both directions. Push work down to a cheaper model to save money, or pull the strongest models up as advisors to make the work better, and not just one:

• Convene a panel. Spin up several strong models in parallel, one from each subscription if you like (say Sol on Codex, Fable on Claude, Hy3 on OpenRouter), to review a plan or a diff at the same time.
• Act only on consensus. Outsourcerer greenlights the work when the panel agrees; a split isn't a coin flip, it's a signal to bring the decision back to you.
• Improve, don't just grade. Ask the advisors to rewrite the weak parts and fold in each other's fixes, not merely flag them.

A second, differently-wired head at the right moment is where the real leverage is. Three of them, agreeing, is even better.

---

One spell, tailored to each model

The models don't think alike, so Outsourcerer doesn't prompt them alike. The right prompting techniques are baked in per capability tier: it frames a task for GPT-5.6 one way, for Fable another, for a genuinely small model a third, each to its strengths. And cheap does not mean dumb: GLM-5.2, Hy3 and DeepSeek are capable tier, frontier capability at a budget price (~Opus-4.8 class), so they get the same high-autonomy prompting as the flagships, not a hand-holding work order. You ask in plain language; the sorcerer translates it into the dialect each engine actually responds to. You get better output from the cheap lane than you'd get by sending it the prompt you'd send Claude. Dial thinking depth per task with --effort.

---

What you can ask it to do

• Offload the grind. Repo mapping, mechanical refactors, running the test suite, big searches. Cheap model does it; your main agent stays the boss.
• Bring your setup along. Hand the job your skills, plugins, and MCP servers so the delegate works with your tools, not a bare model, even on another harness.
• Convene advisors. Have several stronger models review a plan or a diff in parallel, improve it, and only proceed on consensus.
• Run agents in parallel. Fan out a whole multi-agent gauntlet (a QA sweep, a per-module review, N reviewers) across any backend at once with fanout, watch them live, and collect every finding into one file. No 16-session bootstrap tax.
• Generate images. GPT-image first on your Codex plan (keyless, no API credits), then nano-banana or an OpenRouter image model. (Every illustration in this README was rendered exactly this way, by Outsourcerer, keyless, $0 cash.)
• Review anything visual. Screenshots, UI, design, handed to Gemini, keyless on your Antigravity login.
• Find the cheap seats. Ask what's free or cheap right now across your platforms (suggest). It reads each catalog live, so it keeps up as new models drop (a fresh free model on Devin or OpenRouter shows up on its own).
• See what you're saving. The Tab shows real spend vs your frontier model, per lane.

All of it in plain language. The commands underneath are the sorcerer's, not yours.

---

The Tab: a receipt you can actually believe

Most "cost saver" tools show you a number they made up. This one keeps two honest columns, because they aren't the same kind of cost. Cash is real money on OpenRouter/API lanes. Plan limits are your Codex / Claude / Antigravity / Devin subscription windows: no cash, but finite, so we show what you actually burned of them, never a fake "free."

  == The Tab ==  (real output, 2026-07-10)
  runs recorded          : 12
  cash billed (measured)  : $0.002245   REAL per-generation OpenRouter cost, captured on bg runs
  cash lanes, est-only    : 5 run(s)    foreground; run in bg to capture measured $
  on your subscription    : 3 run(s)    $0 cash, spent your ChatGPT / Claude / Antigravity plan limits
    ChatGPT plan usage, 5h window: 6% (resets in 4h 32m) · weekly: 1%
  note: a $0 cash line is NOT "free", subscription lanes spend finite plan rate limits.

That $0.002245 is not an estimate. For OpenRouter lanes the Tab reads the exact per-generation cost back from the provider after the run, because we learned the hard way that the harness's own built-in cost number ran roughly 28× high on cheap lanes. For the subscription lanes it reads the real 5-hour and weekly rate-limit numbers the CLI records after each call. The archmage only gets billed for archmage work, and a no-cash line never pretends a plan lane was free.

---

No wands, just plumbing

Under the robe it's deliberately boring: a self-contained bash script you can read in one sitting. No server, no proxy, no telemetry, nothing resident. It shells out to the CLIs you already have (claude, codex, devin, agy), reads exactly one key from ~/.env when a paid lane needs it, and keeps the ledger in a local JSONL on your machine. The magic is in the routing decisions, not the machinery, which is exactly where you want it.

When it goes sideways: a stalled delegate gets killed by a watchdog and reported with its last progress line, never silently retried against a half-edited tree. Exit codes distinguish done / done-but-unverified / blocked, and "done-but-unverified" is treated as unverified. Before it dispatches, doctor preflights your machine, so if a lane's helper is missing it warns you up front instead of letting a delegated model discover it mid-run.

---

How it compares

	Claude's own subagents	Router / proxy (claude-code-router, LiteLLM)	🧙 Outsourcerer
Shows you the money saved	✗	✗	✅ the Tab, per lane, vs your frontier model
Crosses agents, not just models	Claude only	model-swap inside one harness	Codex · Antigravity · Claude Code · Devin
Carries your skills / plugins / MCP to the delegate	✗	✗	✅ your whole setup, on any harness
Keyless on what you already pay for	Claude sub only	API keys	✅ your existing subscriptions, or the API key of your choice
Advisor panel + consensus	✗	✗	✅ several top-tier models, act only when they agree
Per-model prompting built in	✗	✗	✅ tier-aware wrappers
Setup	zero	proxy + routing config	none, you just talk to it
Server-side routing at scale	✗	✅	✗ this is a local tool, on purpose

---

Handled with gloves (security)

Audited by repo-forensics on every release: 0 critical, 18 high, 1 medium, every finding triaged by name in SECURITY.md. (The 18 "high" are correlation flags for a delegation tool doing exactly what it says: reading one scoped key, calling one provider, supervising background jobs.)

• Keyless by default where your subscription allows, Gemini/Antigravity, GPT-image, and the native Claude / Codex / Devin lanes. Keyless means no API key and no cash, not free: these lanes spend your subscription's finite 5-hour and weekly limits, and the Tab shows how much. Keys are only for the paid OpenRouter lanes.
• Single-key sourcing. Only the one key a lane needs is read from ~/.env, never your whole environment. A budget model never sees a secret it wasn't handed.
• Tells you the lane and tier on every dispatch (and the sandbox posture, read-only / can-write), and asks before anything destructive.
• :free models may train on your prompts. Outsourcerer flags :free lanes as may-train and defaults them to least-privilege, so you decide before sensitive context goes down one.

---

Get it (30 seconds, per host)

Your agent	One-time install
Claude Code	/plugin marketplace add alexgreensh/outsourcerer then /plugin install outsourcerer@outsourcerer
Antigravity	agy plugin import claude-code (or let Outsourcerer's parity mirror it in)
Codex	outsourcerer parity-codex (adds the entry point Codex reads)
Devin	outsourcerer parity (mirrors the skill + your local tools)

Then just talk to it. On first use it runs its own health check and tells you if anything needs installing, with the exact command.

---

Pairs well with

• repo-forensics. The security auditor (27 scanners, 500+ patterns) that vets this plugin, and anything else, before you install it. That green badge up top is its verdict.
• Token Optimizer. Keeps your main agent's context lean while Outsourcerer moves the work off it. Two halves of "spend less, do more." (Its calibrated token math is what the Tab uses to price a run.)

Under the hood, the commands the sorcerer drives for you (you rarely type these)

The model alias picks the lane automatically; --provider is only for the OpenRouter/Devin lanes.

Verb	Does
run / research / edit / yolo	offload a read / sandboxed investigation / code change / approvals-off change
continue / session	multi-turn follow-ups (Devin)
bg / status / watch / result / logs / cancel	background jobs + the liveness watchdog
second-opinion	consensus-gated re-check
image	text-to-image (GPT-image preferred → nano-banana → OpenRouter)
tab / estimate	cost ledger + pre-flight quote
suggest / deals	live cheap & free models available per platform right now
doctor / models	health check + live model list
parity / parity-codex	mirror into Devin/Antigravity; reverse-bridge into Codex

Lanes: hy3/glm-5.2/deepseek-* (OpenRouter, via cc or codex) ·
sol/terra/luna (Codex native) · fable/opus/sonnet/haiku (Claude native) ·
gemini-pro/gemini-flash (Antigravity, keyless) · gpt-image/nano-banana (images).

Add capability to one offload with --with skills=<name> / --with mcp=<name>.

---

Roadmap

More harnesses, coming soon. Today it works across Codex · Antigravity · Claude Code · Devin. Next up:

• Cursor (high priority), the harness most of you asked for first.
• Pi and more harnesses as the frontier expands.
• insourcerer, the reverse bridge, first-class.
• An OpenRouter server-side subagent lane, and worktree-isolated parallel fan-out (basic parallel fanout shipped in v0.2.0).

Issues and PRs welcome.

---