Power your agent
คำสั่งของผู้เผยแพร่เอง ไม่ใช่เอนด์พอยต์ของ PlugMCP
Host a copy with our keys and logs.
npx -y [email protected]Claude Code ไม่มีลิงก์ติดตั้งสำหรับคำสั่งแบบโลคัล ให้คัดลอกคำสั่งด้านล่างแล้วรันในเทอร์มินัล
claude mcp add phonton-cli -- npx -y phonton-cli@0.26.0รันในเทอร์มินัล
{ "mcpServers": { "phonton-cli": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}claude_desktop_config.json
URL แบบรีโมตหรือคำสั่งบนเครื่องของคุณ? การติดตั้งสองแบบต่างกันอย่างไร
รันบน PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
เปิดจากผู้เผยแพร่
npm install -g phonton-cli
phonton doctor # checks your provider key, git, and local tools
phonton # opens the TUI; type a goal and press Enter
No account, no Phonton server. Bring a key from Anthropic, OpenAI, DeepSeek, OpenRouter, Gemini, Groq and others, or run on local models (preview, below).
From the recorded run above (DeepSeek Flash, small JavaScript repo):
Local-first ADE CLI for planned, verified, reviewable code changes.
goal Make parsePort reject ports outside 1-65535 and trailing garbage
plan 1 subtask, verify with the repo's own `node --test`
edit unified diff against the exact current src/port.js
verify patch applies · syntax · tests: 3 pass, 0 fail (was 1 pass, 2 fail)
review receipt: 1 file (+7/-1) · 803 in / 439 out tokens · $0.0008 · 8.2 s · checkpoint #1
remember completion stored locally for the next goal
Nothing lands in your tree until it passes verification, and every run leaves a
receipt you can audit with phonton review latest and phonton why-tokens.
GoalContract with
acceptance criteria and a verify plan before any worker starts.HandoffPacket.Run on your own GPU with no key. Phonton measures what the model can do, asks it only for edits in a format it handled during calibration, runs your checks on a copy, and lands nothing until you apply the result.
phonton models catalog # marks the model for this machine
phonton models setup qwen3.5:9b # runtime, download,
# calibrate edit formats, select
phonton # TUI: goals now run on the local model
phonton goal "Add a count() method to TodoStore" --yes --allow-host-checks
phonton goal --local apply RUN_ID --yes # after reviewing the verified candidate
phonton goal --local rollback RUN_ID --yes # restore the original files
With provider ollama and a calibrated model selected, both the TUI and
phonton goal use the local harness. Every goal pauses for plan approval,
local runs never fall back to a cloud model, and the receipt says whether the
runtime was verified as Phonton-managed.
models setup installs a pinned, hash-checked Ollama runtime that Phonton
starts and verifies, on Windows x64, Linux x64/arm64 and macOS (about 1.5 GB
to download on Windows and Linux, 160 MB on macOS). If an Ollama you
installed already answers on port 11434, Phonton uses it instead but cannot
verify it (an Ollama install can relay cloud models): the TUI plan review says
so and your approval is consent for that goal; headless runs need
phonton goal --local ... --allow-unverified-runtime.
models catalog recommends one model by total memory (the largest GPU, or
half of RAM when that is larger):
Under 4 GB, or without a memory reading, nothing is recommended. The
recommendation is a size-based starting point; calibration decides
whether a model is usable. Any other Ollama tag works with models setup.
In our smoke runs, qwen2.5-coder:3b on a 6 GB laptop GPU finished small single-file edits in a Node repo in 20-30 s. Small models work best on one focused change at a time. Details and limits: docs/local-harness.md.
npm install -g phonton-cli
Without Node: download the binary for your platform (Linux x64, macOS
Apple Silicon, Windows x64) from
GitHub Releases
and put it on your PATH.
From source (needs stable Rust and a C/C++ toolchain):
cargo install --git https://github.com/phonton-dev/phonton-cli phonton-cli --locked --force
# Launch the interactive Ratatui TUI
phonton
# Run a goal non-interactively through plan/edit/verify/review
phonton goal "add input validation to config loading" --yes --allow-host-checks
# Run an exact prompt file, useful for benchmark and CI harnesses
phonton goal --prompt-file prompt.md --yes --allow-host-checks --json
# Verify with your own test command instead of the detected checks
phonton goal "fix the parser" --check '["python","-m","pytest","-q"]' --yes --allow-host-checks
# Preview the task graph and GoalContract without editing files
phonton plan --json "refactor auth layer"
# Audit configuration, provider key, store, trust, and the project's toolchains
phonton doctor --provider
# See where the latest goal spent tokens
phonton why-tokens
# Review the latest receipt, then export typed proof
phonton review latest
phonton proof export --latest --format json
# Inspect MCP capability proposals without invoking tools
phonton mcp capabilities <server-id> --json
Configure providers and the code index in ~/.phonton/config.toml:
Set PHONTON_CONFIG_PATH to an absolute config path outside repositories for
an isolated CLI run. Local-model state stays at its usual location unless
PHONTON_LOCAL_STATE is also set. Provider keys can remain in process
environment variables instead of the isolated file; config-file keys take
precedence when present. The normal config is left untouched.
[provider]
name = "deepseek"
model = "deepseek-flash"
[provider.keys]
deepseek = "sk-deepseek-api-key-here"
anthropic = "sk-ant-api-key-here"
openai = "sk-proj-openai-key-here"
[index]
backend = "local-hnsw"
A goal that keeps failing verification escalates to the provider's larger
models. Add pin_model = true under [provider] to use model on every
tier instead, for example to compare models fairly.
Optional Qdrant code retrieval:
[index]
backend = "qdrant"
qdrant_url = "http://127.0.0.1:6333"
qdrant_collection = "phonton-code"
When a goal misbehaves, PHONTON_LOG=debug (or a narrower filter such as
phonton_worker=debug) appends diagnostics to phonton.log in the Phonton
home: each worker attempt, why verification rejected it, and hunks Phonton
moved to their matching lines.
Phonton is designed for context efficiency and accountable verification, but public comparisons require reproducible evidence: pinned fixtures, exact prompts, tool versions, model/provider names, provider-reported token usage where available, raw logs, final diffs, verification logs, quality review, and handoff evidence.
benchmarks/smoke runs fixed goals on fresh copies of fixtures/ and scores
them with hidden tests the run never sees; any build can be compared the same
way. Method, commands and claim rules: docs/BENCHMARKS.md.
phonton-cli: TUI, headless goal runner, benchmark export, and CLI commands.phonton-types: shared GoalContract, HandoffPacket, PlanGraph, events, and provider types.phonton-planner: goal decomposition, contract generation, and plan graph metadata.phonton-orchestrator: task scheduling, confidence gate, retries, verification, and handoff assembly.phonton-worker: context assembly, provider calls, diff-only output, repair prompts, and MCP flow.phonton-index: local HNSW symbol/code retrieval plus optional Qdrant retrieval.phonton-verify: patch, syntax, decision, Cargo, Node, and browser verification.phonton-memory and phonton-store: local memory facade and SQLite persistence.phonton-sandbox: command and tool execution guardrails.phonton-extensions and phonton-mcp: local extension loading and MCP runtime.phonton-local: hardware detection, managed Ollama runtime, local model install and calibration (preview).Licensed under either of:
At your option.
On a calibrated local route, every goal pauses at its proposed files, exact check arguments, model, runtime origin and budget. Press Y to approve that plan or N / Esc to cancel. Enter, paste and held-key repeats do not approve a plan. This is separate from host-check permission; plan approval never grants access to run project commands on the host. Changes remain in a copy until a verified candidate is explicitly applied.
Local receipts and the run record include reported tokens from rejected and interrupted strategy proposals as well as edit candidates. Only a verified managed runtime counts as local inference; external or unobserved origins are reported as unknown. The TUI records a local result only after its final receipt has been saved.
Local-first ADE CLI for planned, verified, reviewable code changes.
{ "mcpServers": { "phonton-cli": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}เครื่องมือ
ยังไม่มีรายการเครื่องมือ เซิร์ฟเวอร์แบบรีโมตจะได้รายการหลังการตรวจสอบสถานะครั้งแรก
FAQ
คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
Phonton Cli ทำงานแบบโลคัลผ่าน stdio
ตรวจสอบล่าสุดเมื่อ 2026-11-12 Local-first ADE CLI for planned, verified, reviewable code changes.