pulseaiclub/phi

• 15 MB · ~31 ms · sub-agents · anchored sloppy edits · permission gate · progressive MCP · PXB extensions · native diff review & code selection · OpenAI / Anthropic / Gemini

• Docs
• Quick start
• Footprint
• Configuration
• Interactive mode
• Diff review
• Code viewer
• Commands
• Sessions
• Headless mode
• Skills
• Permissions
• Extensions
• MCP
• Tools
• Project layout

Install the latest release (macOS / Linux):

curl -fsSL https://raw.githubusercontent.com/pulseaiclub/phi/main/scripts/install.sh | bash

Windows (PowerShell 5.1+):

irm https://raw.githubusercontent.com/pulseaiclub/phi/main/scripts/install.ps1 | iex

First launch needs a model. Open the config editor (creates ~/.phi layout and writes ~/.phi/config.yaml):

phi config

Or set env vars for a one-off run:

export PHI_MODEL=gpt-4o
export PHI_API_KEY=sk-...

Then start the TUI:

phi

Or build from source (Go 1.26.3+, see go.mod):

make build          # produces ./phi
make install        # build and install into $GOBIN

On first start, phi automatically creates ~/.phi/{bin,skills,hooks,session}. Search tools (fd, rg) download into ~/.phi/bin in the background when missing.

The TUI gives the model four core tools — read, write, edit, and bash — plus grep, find, and ls. The model uses these to fulfill your requests. External HTTP fetch is available via MCP when configured.

Lean is not enough — phi is built to feel instant and stay cheap under load. phi numbers are a stripped release build (CGO_ENABLED=0, -ldflags="-s -w") on macOS arm64. Other harnesses use published Linux PSS / interactive PTY figures.

phi reads ~/.phi/config.yaml (standard YAML). Environment variables override it for one-off runs. phi config opens a full-screen terminal editor for the same file — nothing is written until you save, and the previous file is kept as config.yaml.bak.

phi config keys
  ↑↓ ←→      move / cycle a value        a   add a model
  ⏎          edit or open a picker       d   delete the focused row
  esc        close, or quit              f   fetch model ids from the provider
  ^s         save                        s   save        q  quit

An empty field means "not set", i.e. the value the loader would fill in; the placeholder shows that value.

# ~/.phi/config.yaml
models:
  - name: gpt-4o
    api: OpenAI             # OpenAI | OpenAIResponses | Anthropic | Gemini (empty → OpenAI-compatible)
    api_key: sk-...         # or set PHI_API_KEY
    base_url: https://api.openai.com/v1   # default; PHI_BASE_URL overrides
    context_window: 128000  # optional
    default: true           # the model used at startup; first entry wins if absent
  - name: claude-sonnet-4-20250514
    api: Anthropic          # required — no name/URL guessing
    api_key: sk-ant-...
    base_url: https://api.anthropic.com
    context_window: 200000
  - name: deepseek-flash    # built-in preset: base_url / context / thinking filled in
    api_key: sk-...
  - name: gemini-2.5-flash  # built-in preset (api: Gemini)
    api_key: ...
    think_level: high       # optional: off | minimal | low | medium | high | …

skill_path: ~/.phi/skills # where SKILL.md files are loaded from; "~/…" is expanded
max_rounds: 64            # tool round budget; 0 → engine default (64)

agents:
  enabled: true           # default; set false to disable agent_* sub-agent tools
  models:                 # optional per-role defaults; omit → inherit parent model
    explore: cheap-model
    review: strong-model
    worker: coding-model

permissions:
  mode: interactive       # interactive | readonly | autopilot | headless-strict
  bash:
    default: ask          # ask | allow | deny
    allow:
      - "go test ./..."
    deny:
      - "rm -rf *"

Built-in presets and thinking wire formats: doc/models.md.

Environment overrides:

Security: When PHI_OPTIMIZER is enabled, command history and context are sent to an external TypeSafe service for judging and ranking. Avoid using the optimizer in environments where shell history may contain sensitive information (API keys, tokens, credentials). Provider routing uses the explicit api field (OpenAI / Anthropic / Gemini). See Supported models.

~/.phi/
├── config.yaml   # global configuration
├── bin/          # downloaded search tools (fd, ripgrep)
├── skills/       # SKILL.md skill directories
├── extensions/   # PXB binaries + phi.yaml
├── jobs/         # sub-agent job artifacts (meta, logs, result.md)
└── session/      # persisted sessions, one dir per working directory
    └── <encoded-cwd>/

phi (or phi tui) starts the TUI: a chat transcript on top, an editor at the bottom, and a footer with the current activity. When a newer release is available, the footer shows a hint like 0.2.0 available · phi update.

Assistant output is rendered as Markdown (CommonMark/GFM): headings, emphasis, strikethrough, links, blockquotes, lists, task checkboxes, and tables are styled with the active theme; fenced code blocks get a muted language caption and per-language syntax highlighting. Structural markers (#, `, *) are stripped.

The editor supports:

• @ — fuzzy file mention picker (type @ and start typing a path)
• / — slash command picker (/sessions, /branch, /new, /diff, /code)
• ? — shortcut help picker (lists /, !, @, and key bindings; Esc closes)
• !command — run a shell command locally and stream its output into the transcript (see Commands)
• Ctrl+K — command palette: settings → model / theme / permissions / agents (incl. per-role models), skills, hooks

Themes: Dark (default), Darcula, Pink, and Terminal, switchable from the palette under settings → theme.

/diff is a full-screen git review inside the TUI — read the change, leave line notes, then hand them to the agent without leaving the terminal.

Slash-picker Enter inserts /diff plus a trailing space into the composer; submit to open. Inside the overlay: s side-by-side, i add/edit a note, x delete, a send notes to the agent, ? help, q / Esc close. Notes live in the overlay's memory only — switching diffs drops them and nothing is written to disk.

/code <path>[:line] opens a full-screen source viewer: syntax highlighting, a CJK-aware caret (j/k, h/l, gg/G), in-file search (/, then n/N), :line to jump to a line, and v for lines or p for the paragraph under the caret, then a to hand the selection to the chat input as a path:12-18 reference the model reads. The pane reads files itself and answers to the same deny list as the tool gate, so sensitive paths (~/.ssh, .env, …), binaries and files over 8 MiB are refused with a toast.

Esc closes, or cancels a search back to where / was pressed. The status row shows the path, caret and line count while reading, the selection size while v is active, and the match counter (2/5) while a search is live.

/branch opens the branch picker for the working directory. Two columns: the branch name, and its most recent commit. A ● marks the branch HEAD is on. Rows are ordered where you are, where you came from (git reflog), the other local branches, then remote-tracking branches.

Enter runs git switch on the selected row — off the UI goroutine, so a slow checkout does not freeze the composer. Picking a remote row checks out the local branch that tracks it, the same thing git switch feat does for origin/feat.

/branch <name> skips the picker: it switches to that branch, or creates it from HEAD when no branch carries the name yet. Naming is the whole "new branch" flow — three characters typed beat a form.

Uncommitted work is never touched: git refuses rather than lose it, and its reason is toasted verbatim. Switching is blocked while a reply or command is running, and while a merge, rebase, cherry-pick, revert, or bisect is unfinished.

In the TUI, !command runs locally via bash -c — outside the agent loop. It doesn't count toward agent busy state, and the running command can be cancelled with Esc without touching an in-flight agent turn.

Sessions persist automatically per working directory under ~/.phi/session/<encoded-cwd>/ as JSONL trajectories.

• phi sessions list — list session id, mtime, and preview for the current directory
• /sessions in the TUI — same, in-app
• /new — start a fresh session (new id, empty transcript)
• phi run --session <id> / phi run --continue-last — resume headlessly

phi run -p "fix the failing test in internal/tools"

Runs one agent loop without a TUI. Human logs go to stderr; with --jsonl, machine-readable events go to stdout, one JSON object per line.

Flags:

--tools accepts built-in names such as read,ls,grep. MCP and agent tools still append when configured; the flag only scopes the built-in toolset.

Exit codes: 0 success · 1 runtime/LLM error · 2 max rounds reached · 3 config/usage error.

In the interactive TUI, exhausting the tool-round budget prompts Continue / Stop. Headless phi run has no confirmation UI, so it exits with code 2.

In headless mode, permission ask decisions are denied (there is no approval UI), so readonly-style safety applies without extra flags. For benchmarks that need arbitrary shell (pytest, npm test, …), pass --yolo to skip the permission gate for that run only.

Skills are directories containing a SKILL.md file with YAML frontmatter and a Markdown body. They are loaded from ~/.phi/skills/ (or skill_path / PHI_SKILL_PATH) and injected into the agent's context, letting you give the model reusable procedures:

The Agent Skills locations ~/.agents/skills/ and .agents/skills/ are read too, so a skill pack installed for another agent works here without configuration. Project .agents/skills/ directories are discovered from the working directory upward, stopping at the repository root when there is one. Roots are searched in order — skill_path first, then ~/.agents/skills, then the project directories — and the first root wins when two of them define the same skill name. A leading ~ in skill_path or PHI_SKILL_PATH expands to your home directory.

Security: skill files are workspace content, not trusted input — a project .agents/skills/ is picked up without confirmation. Review skills the way you review AGENTS.md before trusting a checkout.

---
name: My Skill
 description: What this skill does
license: MIT
compatibility: claude, openai
---
Instructions the agent should follow when this skill is relevant.

In the TUI, add skills from the palette (skills → list), then submit the message with the selected skills applied.

Tool execution is gated by a permission policy, so the agent can run read-only by default and ask before anything destructive. Configure it under permissions: in ~/.phi/config.yaml.

Modes:

Per-tool rules: bash.default / bash.allow / bash.deny (exact command prefix matching). Global keys: workspace_only_writes (default true), ask_timeout_sec, and dangerously_allow_all (default false).

In the TUI, an approval dialog replaces the editor with options to approve, deny with feedback, or allow all for the session / for every session. The palette's settings → permissions entry toggles session-wide bypass.

Extensions are native binaries speaking the PXB binary protocol over stdin/stdout (author SDKs: Go github.com/pulseaiclub/phi/ext/go/phi, Rust ext/rust (phi-ext), and TypeScript ext/ts (@pulseaiclub/phi-ext)). They subscribe to tool/session events, register LLM tools, and add slash commands.

go get github.com/pulseaiclub/phi/ext/go@v0.21.0

package main

import (
	"github.com/pulseaiclub/phi/ext/go"
	"github.com/pulseaiclub/phi/ext/go/phi"
)

func main() {
	m := phi.New("hello", "0.1.0")
	m.OnToolCall(func(ev ext.ToolCallEvent) *ext.ToolCallResult {
		// return &ext.ToolCallResult{Block: true, Reason: "..."}
		return nil
	})
	_ = m.Run()
}

Install under ~/.phi/extensions/<name>/ with a phi.yaml pointing at the binary. In the TUI: Ctrl+K → extensions. Disable with PHI_EXTENSIONS=off. Full guide: doc/extensions.md.

Codec throughput on Apple Silicon (release, single-threaded):

The ~10× gap over JSON lines is the protocol (fixed header + tagged fields), not the language — Rust and Go are within noise on the same codec work. Real extension latency is dominated by process spawn and pipe RTT anyway. Re-probe: cargo run --release --example bench in ext/rust.

Configure 100 MCP servers. Pay ~0 schema tokens until you call one.

Most MCP hosts dump every tools/list schema into the model context before you ask a question — browser stacks alone can burn 50k+ tokens. phi does not.

Instead MCP is a URL namespace on the regular tools, and the system prompt lists configured server names (no schemas):

Flow: pick a server from the prompt → ls → read → write. Subprocesses start lazily on first use. Browsing is read-tier; invokes still go through PreHooks → write Gate / Ask → Run → PostHooks.

phi mcp add browsermcp -- npx @browsermcp/mcp@latest
phi mcp doctor
# In the TUI, the model can use configured servers without guessing MCP exists

Config: ~/.phi/mcp.json (project <cwd>/.phi/mcp.json overrides by name). Disable with PHI_MCP=off. Stdio and HTTP in v1.

Full guide: doc/mcp.md.

Sub-agent tools (agent_spawn, agent_wait, …) are on by default. To keep a session lean, disable them in ~/.phi/config.yaml:

agents:
  enabled: false

Or toggle for the current session via the palette: settings → agents. When disabled, those tools are not registered and the model cannot spawn jobs.

Per-role model defaults (optional) under agents.models pick which configured model name each role uses when spawned. Omitted roles inherit the parent session model. Switch for the current session only via settings → agents → models → explore|review|worker (same session-only semantics as settings → model; does not write config.yaml).

Sub-agents themselves use a role (explore default | review | worker):

Default stays explore (no edits). Prefer worker when the task is to implement a scoped change in an isolated context.

Built-in tools the model can call (see internal/tools/):

Sub-agent transcripts live under ~/.phi/jobs/<id>/ and are not injected into the parent context — only the wait/task summary is.

Fast search tools (fd, ripgrep) are downloaded on first startup into ~/.phi/bin when missing.

See Project layout for the source tree map.

See CONTRIBUTING.md for development setup, code style, and commit conventions.