ZacharyZhang-NY
1bb10ef7f1
Root cause of 'connected Claude, still shows Kimi / unknown model': every fetch-plan decision consulted only the primary (Kimi) session and API keys — never the stored subscription-OAuth sessions: - on_auth_changed's wipe guard: a claude-pro-max-only login satisfied 'no session, no keys' → catalog wiped, early return BEFORE the fetch and BEFORE notify_models_updated. The TUI kept an empty picker and the prompt bar rendered 'unknown'. Guard decision extracted into the pure should_wipe_catalog_on_auth_change (matrix-tested); a stored OAuth session now vetoes the wipe, so the fetch runs, the first real catalog reselects the default model (first entry = the connected provider's), and kigi/models/update reaches the client. - Startup prefetch: the arming gate ignored stored OAuth sessions and the prefetch thread passed an empty token map — a claude-only user booted onto the bundled Kimi table until a later refresh. The gate now takes has_stored_oauth and the thread resolves each stored session's bearer (refresh-on-expiry) via a current-thread runtime. - Cache origins: from_config's startup cache load and cache_origin() computed the fetch-plan origin with an empty token map, so a claude-inclusive cached catalog never matched at startup. Both now use presence-only stubs (stored_oauth_token_stubs — names only, no bearers) proven equal to the real-token origin by test. New probes in models_fetch: stored_oauth_platforms / stored_oauth_token_ stubs (sync auth.json scope scan; no AuthManager, no secrets). Verified: kigi-shell 5260 tests green, clippy clean.
Kigi (kigi) 🌘
🕸️ The world's first CLI with built-in Graph Engineering
/graph turns one objective into a dependency graph of
autonomous, self-verifying agent loops — planned, parallelized,
adversarially verified, and merged back, end to end.
Kigi is a coding agent that lives in your terminal. It reads the repo,
writes the patch, runs the tests, and keeps going while you do something else.
Full-screen, headless in CI with -p, or docked in your editor over ACP.
Already paying for Claude Pro/Max, ChatGPT Plus/Pro, GitHub Copilot, or Grok? Sign in and use it. No API key, no second bill. Rather bring your own key? OpenAI, Anthropic, Google, DeepSeek, Groq, Moonshot and two dozen more are wired in.
Installation · Graph engineering · Providers and API keys · Building from source · Coexistence with the official CLI · License
Installation
# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/ZacharyZhang-NY/Kigi-CLI/main/install.sh | bash
# Windows PowerShell
irm https://raw.githubusercontent.com/ZacharyZhang-NY/Kigi-CLI/main/install.ps1 | iex
kigi login # pick a provider, sign in
kigi # go
Single file, no runtime. macOS and Linux on arm64/x86_64, Windows on x86_64,
checksummed against the release's SHA256SUMS. kigi update handles upgrades.
Graph engineering
Every other agent runs a loop: think, act, repeat — one thread, one thing at a
time. /graph runs a dependency graph instead. Work that doesn't block other
work happens at the same time, in separate worktrees, and nothing merges until
something else has tried to tear it apart.
/graph <objective> [--budget <tokens>] # decompose + run fully autonomously
/graph status # node tree, budget, current work
/graph show # box-drawing DAG view
/graph pause | resume [--budget <n>] # halt / continue (budget top-up)
/graph clear # abandon the graph
One command runs the whole thing, start to finish:
- A planner breaks your objective into a dependency DAG, then validates it.
- Independent nodes fan out as parallel workers, each in its own git worktree.
- Every node has to get past an adversarial verifier before it merges back.
- Find something out of scope? Say
DISCOVERED:and the graph replans — append-only, so nothing already agreed on gets rewritten. - Between passes, a topology optimizer drops dependencies that were never real.
- A final node re-checks the whole objective before the graph is allowed to call itself done.
State lives in .kigi/graph.jsonl, next to your code. Close the laptop, come
back tomorrow, /graph resume. A teammate can pick it up from the same file.
On by default. KIGI_GRAPH=0 turns it off; KIGI_GRAPH_CONCURRENCY (default
3) controls how many nodes run at once.
Providers and API keys
29 platforms ship compiled in: 5 you sign into, 24 you hand a key. Nothing is registered at runtime — if it's not in this list, it's not there.
Sign in with a subscription you already pay for. Run kigi login and pick.
Each provider's token is stored under its own key, and one provider's
credentials are never sent to another.
| Platform id | Provider | Sign-in |
|---|---|---|
kimi-code |
Kimi Code (original target) | Subscription OAuth (device code) |
claude-pro-max |
Claude Pro/Max | Subscription OAuth (browser, PKCE) |
openai-codex |
ChatGPT Plus/Pro (Codex) | Subscription OAuth (browser, PKCE) |
github-copilot |
GitHub Copilot | Subscription OAuth (device code) |
xai-grok |
xAI Grok | Subscription OAuth (device code) |
You get whatever models your plan actually serves — the list is fetched at sign-in, not hardcoded. (ChatGPT/Codex is the exception: its backend publishes no model endpoint, so those four are compiled in.)
API-key providers. Export the env var, or drop the key in
~/.kigi/config.toml. Keys are never logged.
| Provider | Platform id | API key env |
|---|---|---|
| Moonshot (moonshot.cn) | moonshot-cn |
KIGI_MOONSHOT_CN_API_KEY (or KIGI_MOONSHOT_API_KEY) |
| Moonshot (moonshot.ai) | moonshot-ai |
KIGI_MOONSHOT_AI_API_KEY (or KIGI_MOONSHOT_API_KEY) |
| OpenAI | openai |
OPENAI_API_KEY |
| Anthropic | anthropic |
ANTHROPIC_API_KEY |
| DeepSeek | deepseek |
DEEPSEEK_API_KEY |
| Groq | groq |
GROQ_API_KEY |
| Mistral | mistral |
MISTRAL_API_KEY |
| Fireworks AI | fireworks |
FIREWORKS_API_KEY |
| Google Gemini | google |
GEMINI_API_KEY |
| OpenRouter | openrouter |
OPENROUTER_API_KEY |
| Together AI | together |
TOGETHER_API_KEY |
| Cerebras | cerebras |
CEREBRAS_API_KEY |
| NVIDIA NIM | nvidia |
NVIDIA_API_KEY |
| Vercel AI Gateway | vercel-ai-gateway |
AI_GATEWAY_API_KEY |
| xAI (Grok) | xai |
XAI_API_KEY |
| Qwen Token Plan | qwen-token-plan |
QWEN_TOKEN_PLAN_API_KEY |
| Qwen Token Plan (China) | qwen-token-plan-cn |
QWEN_TOKEN_PLAN_CN_API_KEY |
| Kimi For Coding | kimi-coding |
KIMI_API_KEY |
| Z.AI | zai |
ZAI_API_KEY |
| Z.AI Coding (China) | zai-coding-cn |
ZAI_CODING_CN_API_KEY |
| Xiaomi MiMo | xiaomi |
XIAOMI_API_KEY |
| Xiaomi Token Plan (China) | xiaomi-token-plan-cn |
XIAOMI_TOKEN_PLAN_CN_API_KEY |
| MiniMax | minimax |
MINIMAX_API_KEY |
| MiniMax (China) | minimax-cn |
MINIMAX_CN_API_KEY |
export OPENAI_API_KEY=sk-...
export XAI_API_KEY=xai-...
# ~/.kigi/config.toml
[platforms.openai]
api_key = "sk-..."
[platforms.xai]
api_key = "xai-..."
Model lists sync on startup. Pick one with /model, set its thinking level
with /effort.
Web search/fetch need a Kimi Code subscription; API-key sessions run
without them, same as the official client.
Building from source
rustup toolchain install 1.97.0
cargo build --profile release-dist -p kigi-bin
./target/release-dist/kigi --version
protoc is invoked through the vendored dotslash
launcher at bin/protoc; install dotslash (brew install dotslash or
cargo install dotslash) if it is not already on your PATH.
Coexistence with the official Kimi CLI
Kigi started as an unofficial Kimi Code CLI — a community fork of
xai-org/grok-build, not affiliated
with Moonshot AI or xAI. It keeps its own binary, its own ~/.kigi, its own
keyring entry, and its own KIGI_* env vars, and never touches what the
official kimi CLI installed. kigi import-kimi copies your old config over
once, read-only.
Zero telemetry. It talks to the APIs you configured, GitHub Releases, and your own MCP servers. Nothing else.
License
Apache-2.0. See LICENSE, NOTICE, and
THIRD-PARTY-NOTICES. Code ported from
openai/codex and sst/opencode is documented in
crates/codegen/kigi-tools/THIRD_PARTY_NOTICES.md.
Kigi is based on Grok Build Open Source; the --version output carries the
attribution.