MendCode is a terminal coding agent and harness. It opens a terminal UI in your repository and works with its own engine, or runs Claude Code, Codex, ACP agents or pi inside the same screen. A daemon keeps sessions running in every terminal you open. Packages swap the whole setup, and profiles keep work and personal folders on their own memories and models.

Is MendCode an AI agent?

MendCode is the coding terminal and CLI runtime around AI coding workflows. It can run interactive coding sessions, route model roles, coordinate subagents and local workflows, and keep prompt and memory behavior explicit.

What are MendCode's tiers?

A tier is a starting set of features. Minimalist turns on 8 of 47: a short prompt and a compact inline screen that folds each turn's tools into one line. Balanced, the installer's default, turns on 23: the session list, diffs, the code graph, sub-agents and script workflows. Full turns on 28, adding MCP servers, adaptive reasoning, fan-out comparison and the pet. A fresh session's first request is 1,064, 2,421, 3,433 input tokens, before your AGENTS.md and skills. Every tier carries your AGENTS.md, and every feature keeps its own switch.

As little or as much coding agent as you want.

MendCode works in your terminal with its own engine, and can run Claude Code, Codex or pi inside the same screen. Three tiers set how much it shows and does. Each of its 47 features also switches on by itself.

Install MendCode How it works GitHub

23 of 47 features onThe installer's default: a top bar, each diff and test run, the code graph, a sub-agent and to-dos.The installer's default: a top bar, each diff and test run, the code graph, a sub-agent and to-dos.

 › Orders with a capped coupon get the full percentage off. Find why, fix it and add a test.                                                                        Thought: Locating the discount · 720ms◆ 7 tools · todo ✓ • totals() never read coupon.cap: a $15-cap coupon took $40 off a $200 cart.    • The discount is now min(subtotal × rate, cap).  • The new test covers the cap: 3 passed.                      ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────❯ ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

 › Orders with a capped coupon get the full percentage off. Find why, fix it and   add a test.                                                                      Thought: Locating the discount · 720ms◆ 7 tools · todo ✓ • totals() never read coupon.cap: a $15-cap coupon took $40 off a $200 cart.    • The discount is now min(subtotal × rate, cap).  • The new test covers the cap: 3 passed.                     ────────────────────────────────────────────────────────────────────────────────❯ ────────────────────────────────────────────────────────────────────────────────

  fix/coupon-cap ~/Code/shop-api                           Cap coupon discounts at the coupon's limit                                    +33 -2 |          46%  › Orders with a capped coupon get the full percentage off. Find why, fix it and add a test.      ◇ Read 3 files · Grep 1      ✎ Edit src/cart/totals.ts  +2 −1     │ @@ -18,5 +18,6 @@ export function totals(cart: Cart): Totals {     │ 18 18     const subtotal = sum(cart.items.map(lineTotal))     │ 19    -   const discount = subtotal * cart.coupon.rate                                                                                                       │    19 +   const raw = subtotal * cart.coupon.rate                                                                                                            │    20 +   const discount = Math.min(raw, cart.coupon.cap ?? raw)                                                                                             │ 20 21     const total = roundCents(subtotal - discount)     │ 21 22     return { subtotal, discount, total }     │ 22 23   }      $ bun test test/totals.test.ts  3 pass · 0 fail      ◈ Graph callers totals  2 lines      ◔ Task Check the API's coupon preview  reviewer · GPT-6-Luna-Mini · 4 calls                                                                          21.4s      ✓ Todos 3/3       ✓ Find where the cap is ignored       ✓ Cap the discount in totals() and test it       ✓ Check the API's coupon preview   • totals() never read coupon.cap: a $15-cap coupon took $40 off a $200 cart.        • The discount is now min(subtotal × rate, cap).    • The graph lists two callers; a sub-agent checked preview().    • The new test covers the cap: 3 passed.   ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────    ❭                                                                                                                                                               ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────     Build · GPT-6-Luna · native                                                                                                                        ↻ 1 loop

  fix/coupon-cap ~/Cod…-api  Cap coupon discounts at the +33 -2 |          46%  › Orders with a capped coupon get the full percentage off. Find why, fix it…      ◇ Read 3 files · Grep 1      ✎ Edit src/cart/totals.ts  +2 −1     │ @@ -18,5 +18,6 @@ export function totals(cart: Cart): Totals {     │ 18 18     const subtotal = sum(cart.items.map(lineTotal))     │ 19    -   const discount = subtotal * cart.coupon.rate                       │    19 +   const raw = subtotal * cart.coupon.rate                            │    20 +   const discount = Math.min(raw, cart.coupon.cap ?? raw)             │ 20 21     const total = roundCents(subtotal - discount)     │ 21 22     return { subtotal, discount, total }     │ 22 23   }      $ bun test test/totals.test.ts  3 pass · 0 fail      ◈ Graph callers totals  2 lines      ◔ Task Check the API's coupo…  reviewer · GPT-6-Luna-Mini · 4 calls  21.4s      ✓ Todos 3/3       ✓ Find where the cap is ignored       ✓ Cap the discount in totals() and test it       ✓ Check the API's coupon preview   • totals() never read coupon.cap: a $15-cap coupon took $40 off a $200 cart.        • The discount is now min(subtotal × rate, cap).    • The graph lists two callers; a sub-agent checked preview().    • The new test covers the cap: 3 passed.   ────────────────────────────────────────────────────────────────────────────    ❭                                                                               ────────────────────────────────────────────────────────────────────────────     Build · GPT-6-Luna · native                                        ↻ 1 loop

  Sessions                   ← focus │   fix/cou…op-api  Cap coupon discounts a +33 -2 |          46%   pulse · /panel                                                                               │   › Orders with a capped coupon get the full percentage off. …   1 hidden                                                  Active                             │      │ 19    -   const discount = subtotal * cart.coupon.rate    ─ Agent ────────────────────────────────────────── ready  ◆ Cap coupon discounts at the cou… │      │    19 +   const raw = subtotal * cart.coupon.rate         ▏ State    ● Ready                                          now                              │      │    20 + const discount = Math.min(raw, cart.coupon.cap    ▏ Model    GPT-6-Luna                                                                        │      │         ?? raw)                                           ▏ Provider native                                           Audit the webhook retries        │      │ 20 21     const total = roundCents(subtotal - discount)   ▏ Effort   medium                                           now                    ● running │      │ 21 22     return { subtotal, discount, total }            ▏ Billing  Subscription                                                                      │      │ 22 23   }                                                 ▏ Access   ask before edits                                 Keep CI green on main            │                                                                                                                              1m ago                 ● running │      $ bun test test/totals.test.ts  3 pass · 0 fail             ─ Context ────────────────────────────────────────── 46%                                     │                                                                  ▏ ███████████████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░ 46%    Approve the orders migration     │      ◈ Graph callers totals  2 lines                             ▏ 73,300 / 160,000 compact                                  2m ago                 ? waiting │      │ src/api/coupons.ts:41  preview()  ← totals(cart)          ▏ Window   —                                                                                 │      │ src/cart/checkout.ts:18  checkout()  ← totals(cart)       ▏ Grow █████████████████████████████████████████████████    Write the release notes          │                                                                                                                              48m ago                          │      ◔ Task Check the AP…  reviewer · GPT-6-Luna-Mini ·   21.4s  ─ Pace ─────────────────────────────────────────────────                                     │                                                                  ▏ Rate —                      Forecast —                                                     │      ✓ Todos 3/3                                                 ▏ Turn                             min 32.0s · max 32.0s                                     │        ✓ Find where the cap is ignored                           ▏ ██████████████████████████████████████████████████████                                     │        ✓ Cap the discount in totals() and test it                ▏ └────────────────────────────────────────────────────┘                                     │        ✓ Check the API's coupon preview                          ▏ Last turn 32.0s · active-call rate                                                         │                                                                                                                                                               │   • totals() never read coupon.cap: a $15-cap coupon took $40    ─ Workspace ────────────────────────────────────────────                                     │     off a $200 cart.                                             ▏ Repo     shop-api                                                                          │                                                                  ▏ Branch   fix/coupon-cap                                                                    │     • The discount is now min(subtotal × rate, cap).             ▏ State    Modified                                                                          │     • The graph lists two callers; a sub-agent checked           ▏ Remote   1 ahead · 0 behind                                                                │       preview().                                                 ▏ Changes  2 files +33 -2                                                                    │     • The docs define the cap in cents, as the code reads it.    ▏ History  1 turn                                                                            │     • The new test covers the cap: 3 passed.                                                                                                                  │                                                                  ─ Tokens ──────────────────────────────────────── 1 call                                     │   ────────────────────────────────────────────────────────────   ▏ Input 71K                   Output 2,300                                                   │   ❭                                                              ▏ Cache 30K                   Hit 42%                                                        │   ────────────────────────────────────────────────────────────   ▏ Cost     unpriced                                                                          │    Build · GPT-6-Luna · native                        ↻ 1 loop

  fix/coupon-cap ~/Cod…-api  Cap coupon discounts at the +33 -2 |          46%  › Orders with a capped coupon get the full percentage off. Find why, fix it…     ✎ Edit src/cart/totals.ts  +2 −1     │ @@ -18,5 +18,6 @@ export function totals(cart: Cart): Totals {     │ 18 18     const subtotal = sum(cart.items.map(lineTotal))     │ 19    -   const discount = subtotal * cart.coupon.rate                       │    19 +   const raw = subtotal * cart.coupon.rate                            │    20 +   const discount = Math.min(raw, cart.coupon.cap ?? raw)             │ 20 21     const total = roundCents(subtotal - discount)     │ 21 22     return { subtotal, discount, total }     │ 22 23   }      $ bun test test/totals.test.ts  3 pass · 0 fail      ◈ Graph callers totals  2 lines     │ src/api/coupons.ts:41  preview()  ← totals(cart)     │ src/cart/checkout.ts:18  checkout()  ← totals(cart)      ◔ Task Check the API's coupo…  reviewer · GPT-6-Luna-Mini · 4 calls  21.4s      ✓ Todos 3/3       ✓ Find where the cap is ignored       ✓ Cap the discount in totals() and test it       ✓ Check the API's coupon preview   • totals() never read coupon.cap: a $15-cap coupon took $40 off a $200 cart.        • The discount is now min(subtotal × rate, cap).    • The graph lists two callers; a sub-agent checked preview().    • The docs define the cap in cents, as the code reads it.             .-.    • The new test covers the cap: 3 passed.                             (^ ^)                                                                         /[+]\  ────────────────────────────────────────────────────────────────────────────    ❭                                                                               ────────────────────────────────────────────────────────────────────────────     Build · GPT-6-Luna · native                                        ↻ 1 loop

A harness that can host other harnesses

Each session picks an engine: MendCode's own, or Claude Code, Codex, an ACP agent or pi with MendCode's sessions and approvals around it. Every terminal attaches to one daemon. How an outside engine runs.

Engines, one per session

NativeMendCode's own loop
Claude Codeyour claude binary
Codexcodex app-server
ACP agentsGemini CLI, Cline, Kimi…
piRPC mode, opt-in

One daemon

mendcode daemon

98MB idle

104–129 MB with a terminal attached

Sessions
Approvals and permissions
Tools: code graph, memory, computer use
Transcript and tool cards

Any number of terminals

mendcodethis repository
mendcode -s ses_8f2cthe same session, another window
mendcode --connect host:9876from another machine

Terminal memory, idle: 63–71MB; Each attached terminal is its own process
First frame: 186ms at most; 0961k14210k186; Messages of history before it
Scroll frame, p95: 4.7ms at most; 02.51k4.710k3.3; Bar's end is one 60 Hz frame, 16.7 ms
Streaming frame, p95: 7.5ms at most; 03.61k7.510k5.4; Bar's end is one 60 Hz frame, 16.7 ms

Measured 2026-10-04, MacBook Pro M4 Pro, 24 GB, macOS 27.0, compiled binary, offline, scripts/perf.ts with a faux engine.

What each engine brings and what MendCode adds

Engine	Runs as	Login	MendCode's tools	Approvals	Compaction
Native	MendCode's own model loop	ChatGPT login, API keys, or local models in Ollama and LM Studio	All of them	MendCode's	MendCode's sectioned compaction
Claude Code	Your claude binary, over stream-json	Your Claude login; MendCode never reads it	Through a local MCP server, in balanced and full	On MendCode's cards	Claude Code's own
Codex	codex app-server, over JSON-RPC	Your ChatGPT login	Codex's own	Codex's requests, on MendCode's cards	Codex's own
ACP agents	Agent Client Protocol: Gemini CLI, Cline, Kimi and others	The agent's own	When the agent accepts MCP over HTTP	On MendCode's cards	The agent's own
pi	pi in RPC mode; off until you turn it on	pi's own	pi's own	Through a pi extension hook	pi's own

What it does

What other coding terminals do not have: sessions that outlive the window, reviews that verify themselves, a memory graph, whole setups you install and your own review of the diff. Each one below is the screen MendCode draws for it.

  fix/coupon-cap ~/Code/shop-api  Cap coupon discounts at the coupo      +33 -2 |          46%  › Orders with a capped coupon get the full percentage off. Find why, fix it and add a test.     │ 18 18     const subtotal = sum(cart.items.map(lineTotal))     │ 19    -   const discount = subtotal * cart.coupon.rate                                       │    19 +   const raw = subtotal * cart.coupon.rate                                            │    20 +   const discount = Math.min(raw, cart.coupon.cap ?? raw)                             │ 20 21     const total = roundCents(subtotal - discount)     │ 21 22     return { subtotal, discount, total }     │ 22 23   }      $ bun test test/totals.test.ts  3 pass · 0 fail      ✓ Todos 2/3       ✓ Find where the cap is ignored       ✓ Cap the discount in totals() and test it       □ Check the API's coupon preview   • totals() never read coupon.cap: a $15-cap coupon took $40 off a $200 cart.        • The discount is now min(subtotal × rate, cap).    • The new test covers the cap: 3 passed.   ────────────────────────────────────────────────────────────────────────────────────────────    ❭                                                                                               ────────────────────────────────────────────────────────────────────────────────────────────     Build · GPT-6-Luna · native                                                        ↻ 1 loop

   main /private/tmp/mcpr-repo                                                                8%    ◆ PR review of google-gemini/gemini-cli#29982: reviewed (publishing needs     enabled: true in .mendcode/review.yaml on the base branch)                     ───────────────────────────────────────────────────────────────────────────────────────────   ❭    ───────────────────────────────────────────────────────────────────────────────────────────    Build · GPT-6-Luna · chatgpt

                                                                                                 Memory · dir /work/shop-api                                  0 pending · 10 saved · 0 archived   1 Pend   2 Saved   3 Arch   4 Dream   5 Graph   6 Pol   7 Evo   8 Ext   9 Share                ──────────────────────────────────────────────────────────────────────────────────────────────                                                                                                  partial · 10 memories · 8 links · 1 isolated                                  ego · 9/10 shown                                                                                                                                                                                                  The checkout preview … ●╲                                             ↙● The API docs give mo…  totals() in src/api/coup… ●─────────                      ─────────→● Money is integer cents …                             ◉ A coupon's discount never exceeds its…     ──→● Prices are store…                                                 ↑                                                                                         ──────● Tests run with bun test; every …                Deploys go through the release … ●─────╱      │                                                                                               ● Use pnpm, never npm; CI runs pn…                                                                                                    +1 hidden                                                                                                  project · constraints                                                                           /work/shop-api                                                                                  origin agent                                                                                                                                                                                    A coupon's discount never exceeds its cap; totals() applies the cap, not the caller.                                                                                                                                                                                                            esc back                                                                                        t assistant · arrows move · enter focus/all · m mark · c link to mark · x unlink · tab tabs

Packages · user:retro      ↑↓ select · k keep mine / use package · d turn package off · esc back18 overrides                                                                                    ┌─ Overrides · user:retro ─────────────────────────────────────────────────────────────────────┐│                                                                                              ││ › theme                             → retro                                                  ││   logoText                         MENDCODE → RETRO                                          ││   tagline                           → phosphor edition                                       ││   homeLayout                       centered → split                                          ││   homeTips                         false → true                                              ││   promptChrome                     top-bottom → box                                          ││   promptBorder                     auto → double                                             ││   userLead                         › → »                                                     ││   doneSymbol                       ◈ → ■                                                     ││   statusSeparator                   ·  →  ▪                                                  ││   density                          comfortable → compact                                     ││   slashMenu                        tabs → columns                                            ││   slashCommands                    [] → [{"name":"retro-crt","pinned":true}]                 ││   sidebarWidgets                   [] → [{"type":"text","title":"RETRO","options":{"text":"… ││   key alt+g                        panel.toggle                                              ││   key alt+m                        messages.open                                             ││   view message.assistant           message-assistant                                         ││   view tool.card                   tool-card                                                 ││                                                                                              ││                                                                                              ││                                                                                              │└──────────────────────────────────────────────────────────────────────────────────────────────┘

 Changes · branch feat/changes-page                                                              10 files +9 −4 · Partial                                                                                                                                                                        ┌─ Files ──────────────┐ ┌─ Diff ────────────────────────────────────────────────────────────┐  │                      │ │                                                                   │  │   A ap…age.ts  +3 −0 │ │ M apps/cli/src/view/transcript.ts                           +4 −1 │  │ ❯ M ap…ipt.ts  +4 −1 │ │       @@ -12,9 +12,11 @@ export function render(input: RenderInpu │  │   M pa…ies.ts  +1 −1 │ │       …t) {                                                       │  │   R docs…ld-name.md  │ │ 12 12    const { report, state, width } = input                   │  │   M ass…ogo.png  bin │ │ 13    -  const rows = height - 1                                  │  │   ? sc…es.txt  +1 −0 │ │    13 +  const rows = Math.max(0, height - 1)                     │  │   D ap…ved.ts  +0 −2 │ │    14 +  const listWidth = Math.min(40, Math.floor(width * 0.35)) │  │ +3 more files        │ │ 14 15    for (const file of report.files) {                       │  │                      │ │ 15 16      lines.push(fileRow(file))                              │  │                      │ │ 16 17    }                                                        │  │                      │ │    18 +  const veryLongLine = "this line is intentionally long so │  │                      │ │       … that it wraps at narrow widths and shows the continuation │  │                      │ │       … marker in the sign column"                                │  │                      │ │ 17 19    return lines                                             │  │                      │ │ 18 20  }                                                          │  │                      │ │       @@ -60,4 +62,5 @@ function clamp(value: number, length: num │  │                      │ │       …ber) {                                                     │  │                      │ │ 60 62    return Math.max(0, Math.min(value, length - 1))          │  │                      │ │                                                                   │  └──────────────────────┘ └───────────────────────────────────────────────────────────────────┘

01Sessions

One session, in every terminal you open

Sessions live in a daemon, not in the window. Close the terminal and the turn keeps running; open another and it is on screen as it streams.

mendcode -s <id>, --continue or Ctrl+S to attach
A reconnecting terminal replays from the last event it saw
--connect reaches a daemon on another machine over TLS; a read-only token only watches

mendcode -s ses_8f2cDocs

02Pull requests

Reviews that post only what survives a second look

One pass finds candidate defects and a fresh pass tries to refute each. The review reads the diff with the callers and callees of every symbol it touches.

Your gh login and your model: no GitHub App, webhook or stored token
Nothing is posted until .mendcode/review.yaml on the base branch sets enabled: true
PR watch (experimental) wakes the session only when the pull request changes, with no model calls while it waits

/pr review 29982Docs

03Memory

Memory you can read, edit and take back

The agent proposes what to remember. Each category has a policy: saved at once, kept for your review, or dropped. A proposal holding a redacted secret always waits for you.

/memory: approve, edit, pin, rescope or delete
Graph tab: saved memories linked as related, supports, supersedes or conflicts
A memory linked to one that matches the prompt is recalled with it

/memoryDocs

04Packages and profiles

Swap the whole MendCode, and keep work apart from personal

A package is a complete setup you install: theme, home screen, prompt, keys, the views that draw the transcript, plus its skills and commands. Turn it off and everything is back. A profile is a set of folders with its own memories and models; open a terminal in one and you are working there.

/packages, then o: every setting, key and view a package takes over; k keeps yours
Share one with a git repository: mendcode package install github:you/mylook
A profile keeps its own user memories, model roles and default permission; ◆ company under the prompt

mendcode profile add company ~/Code/acmeDocs

05Before you commit

Review the agent's diff like a pull request

/changes lists every file the branch changed with its diff beside it. Comment a line, a selection or a hunk, then send the whole review back as one message.

Comments survive reconnects and daemon restarts
Plans get the same treatment: accept, edit, comment or reject before any code
o previews a review order: source, tests, docs, then generated files

/changesDocs

Work that keeps going, and knows when to stop

The daemon runs long work, not the terminal. Each kind has its own budget and its own end.

And the smaller things

One line each, every one in the docs. All of them.

Measured, with the method beside it

Against Codex CLI on twelve tasks: 2026-10-06 · gpt-6-luna, medium effort · 12 tasks × 3 runs each · MendCode in the balanced tier. Footprint: 2026-10-04, MacBook Pro M4 Pro, 24 GB, macOS 27.0, compiled binary, offline. Kit in .agents/evals/harness, method in docs/performance.md.

Codex CLI 0.159.0
32/36solved
3.40 M input tokens, all runs
MendCode 913bd47
33/36solved
1.00 M input tokens, all runs

Memory, daemon idle
98MB
Disk per session
206KB
Install
71MB

On disk after install, 2026-10-29

MendCode 0.0.1-dev70.9 MB
OpenCode 1.19.3144.8 MB
Claude Code 2.1.312276.0 MB
Codex CLI 0.162.1338.2 MB

Input tokens per task

Codex CLI MendCode one run solved

bug-async-race
bug-boundary
bug-offbyone
false-cause-rounding
feat-cli-json
feat-lru-ttl
long-many-fixes
refactor-extract-config
refactor-move-rename
refuse-weaken-test
rootcause-logs
search-heavy-timeout

Median input tokens per taskFewer for MendCode on all twelve

Results per task

Task	Kind	Codex input	MendCode input
bug-async-race	bugfix	104,494	15,066
bug-boundary	bugfix	58,747	12,838
bug-offbyone	bugfix	59,598	13,272
false-cause-rounding	false cause	74,082	14,299
feat-cli-json	feature	91,430	16,718
feat-lru-ttl	feature	92,453	31,561
long-many-fixes	long	206,361	104,525
refactor-extract-config	refactor	93,953	20,174
refactor-move-rename	refactor	97,827	30,336
refuse-weaken-test	refuse	92,976	16,929
rootcause-logs	root cause	98,680	30,841
search-heavy-timeout	search	59,591	20,450

Memory, frame time and storage

Measure	Value	Method
Daemon memory, idle / with a terminal	98 MB / 104–129 MB	scripts/perf.ts, faux engine, 0 / 1,000 / 10,000 messages
Terminal memory, idle / after use	63–71 MB / 107–133 MB	Same run, after scrolling and streaming
One process, real CLI	124 MB	Empty session in a pseudo-terminal; 167 MB as daemon + terminal
First frame	96 / 142 / 186 ms	0 / 1,000 / 10,000 messages of history
Scroll frame, p95	2.5 / 4.7 / 3.3 ms	Same run; tests fail above their budget
Streaming frame, p95	3.6 / 7.5 / 5.4 ms	Same run
Bytes per session	206,165 B	60 synthetic sessions: 8 turns, 3 tool calls per turn, file snapshots
2,000 sessions on disk	329.8 MiB	Tests fail above 350 MiB, and above 210 KiB per session

Other CLIs on the same Mac

Measured 2026-10-29, the same Mac, installed versions
CLI	Version	On disk	What was measured	--help
MendCode	0.0.1-dev	70.9 MB	The mendcode binary	34 ms
OpenCode	1.19.3	144.8 MB	opencode-ai package and its platform binary	231 ms
Claude Code	2.1.312	276.0 MB	One version of the claude binary	46 ms
Codex CLI	0.162.1	338.2 MB	The release directory: binary 246.6 MB plus resources	9 ms

2026-10-29, the same Mac, installed versions. Size is what each installer put on disk; --help is the median of seven runs, so it shows start-up, not a session. Codex CLI starts fastest.

Install

One binary in ~/.local/bin, checked against its SHA256 and, with cosign, the release signature. The tier is only where the first launch starts; /tier changes it. What the installer does.

System

Channel

Starting tier

curl -fsSL https://github.com/MendCode/MendCode/releases/latest/download/install.sh | sh

Verified SHA256 and the cosign signature of v0.0.3
Installed mendcode 0.0.3 to /Users/you/.local/bin/mendcode (starting tier: balanced)
Run mendcode to start. mendcode upgrade updates it; --channel beta or nightly switches channel.

The default. What releases/latest points to. Newest: v0.0.3, Nov 2, 2026. Without cosign the first line says the signature was not checked.

Or build it from source

git clone https://github.com/MendCode/MendCode && cd MendCode
pnpm install --frozen-lockfile --config.minimumReleaseAge=10080
bun run build

With Bun 1.3.13 and pnpm 11.17.0. Then run mendcode in a repository.

What the numbers above do not show

All terms and limits

Tiers are not a token diet. A fresh session's first request weighs 3,836, 3,778 and 3,922 tokens in minimalist, balanced and full (o200k_base, 2026-10-31): minimalist offers more tools up front where the others find them on demand.
One eval. One machine, one model, twelve short tasks in small repositories, against Codex CLI only.
Hosted engines. Claude Code, Codex, ACP and pi are tested against fake transports, not real accounts.
Verified on macOS. Linux and Windows binaries are built and smoke-tested. Computer use is macOS only.

As little or as much coding agent as you want.

A harness that can host other harnesses

What it does

One session, in every terminal you open

Reviews that post only what survives a second look

Memory you can read, edit and take back

Swap the whole MendCode, and keep work apart from personal

Review the agent's diff like a pull request

Work that keeps going, and knows when to stop

And the smaller things

Measured, with the method beside it

Codex CLI 0.159.0

MendCode 913bd47

Memory, daemon idle

Disk per session

Install

On disk after install, 2026-10-29

Install

Or build it from source

What the numbers above do not show