NEW Workflow v2 · the multi-agent verification layer is live
75+ models 4 modes 3 personalities parallel sub-agents $50/mo flat

product Defcode

The terminal that works in parallel.

Defcode is an agentic CLI that plans, builds, designs and reviews — spawning sub-agents that work side by side while a live rail shows every task and every cent.

free tier · no card · $50/mo for everything

defcode · ~/api build engineer
read indexed 214 files · AGENTS.md conventions applied 1.2s
plan scoped 4 steps — middleware, config, tests, docs 4 steps
spawn 3 sub-agents → engineer ×2 · writer ×1 parallel
edit src/middleware/rateLimit.ts +118 −6
edit src/config/limits.ts +34 −0
test vitest run — middleware, integration 42 passed
Ready for review · /share to send the session

usage

48.2k tokens
context 38% $0.41

tasks · parallel

rate-limit API

engineer

write changelog

writer

audit deps

researcher

tab — switch mode
3 tasks · 42 tests · $0.41 agents working… claude-fable-5 · /undo /redo /share

01 · Modes

Four modes. One keystroke apart.

One agent, four postures. Hit Tab and Defcode changes how it behaves — from read-only planning to shipping code — without losing the thread.

Tab cycles modes in the TUI
defcode · ~/api build
wire stripe webhooks into the billing service
edit src/billing/webhooks.ts +96 −12
test 18 passed · 0 failed
done — 2 files changed, reviewed gates: 3
how would you migrate sessions to Redis?
read 31 files · auth, middleware, deploy
plan 5 steps · est. 40 min · 2 risks flagged
plan ready — approve to enter build mode
design the onboarding flow, 3 steps max
map flow: signup → connect → first-run
tree <Onboarding> · 7 components, 2 shared
spec written to docs/onboarding.md
review the diff against main
scan 14 files · 412 additions · 88 deletions
flag auth.ts:84 — token expiry uses < not <=
2 findings · 0 nits · fix patch offered

Edits files, runs commands, fixes its own test failures. Every destructive action passes a permission gate first.

Walks the repo, maps dependencies and proposes a scoped, step-by-step plan. No file changes until you approve.

Turns a rough spec into component trees, UI flows and architecture sketches — before a line of code exists.

Walks your diff or PR line by line. Flags bugs, security holes and dead code as one-line findings with fixes.

02 · Personalities & sub-agents

One CLI. A whole team inside.

Pick who shows up to work. Each personality spawns its own sub-agents, so the engineer can refactor while the writer drafts the changelog and the researcher audits your dependencies — all in the same session.

defcode

your session · one prompt

engineer

sub-agents: 3 running idle · spawn on demand

writer

sub-agents: 3 running idle · spawn on demand

researcher

sub-agents: 3 running idle · spawn on demand

personality · 01

Engineer

Reads the codebase like a senior hire. Plans, builds, tests and reviews — the personality you ship with.

personality · 02

Writer

Changelogs, PR descriptions, docs and release notes — written from the actual diff, not from memory.

personality · 03

Researcher

Crawls codebases, docs and the web. Comes back with a brief, sources attached, before anyone writes code.

every sub-agent appears in the task rail the moment it spawns — watch the work, not the spinner

03 · Under the hood

Everything you'd build if you had a year.

The plumbing serious tools are built on — language servers for real code intelligence, MCP for your stack, checkpoints for fearless edits, and privacy that holds up to a security review.

LSP under the hood

Real symbols, references and diagnostics from language servers — not regex guesses over text.

Any model, your call

75+ providers. Frontier, open-source or local — switch mid-session, bring your own keys or use ours.

AGENTS.md rules

Your conventions, written once. Every agent and sub-agent reads them before touching a file.

MCP servers

Plug external tools straight into the agent — databases, browsers, internal APIs, anything with an MCP server.

Share a session

One link sends the whole conversation — prompts, diffs, decisions — for teammates to replay.

Undo anything

Every change is checkpointed. /undo rolls the repo back, /redo brings it forward. No fear, no stash.

Themes & keybinds

A TUI that feels like yours — themes, layouts and every keybind fully remappable.

Permission gates

Nothing destructive runs without explicit approval. Allowlist the safe stuff, gate the rest.

Private by default

Your code is never retained and never trained on. Point it at a local model and nothing leaves the machine.

04 · Pricing

$50. Not $200.

The leading terminal agent tops out at $200 a month. Defcode does more — more modes, more models, more visibility — for a quarter of that. Flat.

compare Defcode Claude Code Max
Price per month $50 flat $200
Modes Build · Plan · Design · Review Default · Plan
Personalities Engineer · Writer · Researcher Not available
Usage & task rail in the TUI Built in, always on Not available
Model choice 75+ providers, local models Single vendor
Parallel sub-agents Unlimited, live in the rail Limited
Session share links Included Not available

Defcode Pro

$50 / month, flat

no seats math · no token anxiety

  • Every mode, every personality, unlimited sub-agents
  • Usage rail with live spend — never a surprise bill
  • 75+ models included, local models free forever
  • Free tier to start — no card, no trial clock
Start free S

upgrade when the free tier feels small

defcode · get started

Open a terminal. Hire a team.

One install command, your repo, and a rail full of agents already working. The free tier needs no card.

macOS · Linux · Windows (WSL) · works with any git repo