Argus-Pi

Argus AI Team's dedicated Pi-based inference and execution harness for Argus.

This is a maintained fork of earendil-works/pi, not a replacement for Argus's orchestration. We optimize the Pi layer: model requests, tool execution, context and session handling, reliability, and task handoffs. Argus keeps ownership of goals, role authority, review and durable mission state.

Current downstream behavior

• The argus harness profile is the default: follow the assigned task and role instead of assuming every invocation is a coding task.
• Keep tool descriptions, Skills, project context, explicit system prompts and decision formats intact. The profile does not grant permissions or add a sandbox.
• Preserve the pi CLI, .pi configuration and existing provider authentication. argus-pi is an additional executable name for the same CLI. Argus-mode JSON output distinguishes recoverable attempt diagnostics from terminal failure, preserving usage and provider-turn accounting; stock, RPC and SDK retain upstream provider-retry event shapes.
• Local Bash uses pipefail so failed experiments, compilers or proof checkers remain failures when their output is piped through tee. This does not enable set -e, retry commands, or change custom remote operations or PowerShell.
• Signal-terminated commands retain partial output and an observable signal. A missing exit code is not treated as success. User/RPC shell records preserve this status in session history, model context, terminal display and HTML export.
• Set PI_HARNESS_PROFILE=stock to restore the upstream default system prompt and local Bash pipeline behavior. Both profiles report terminal text/JSON failures with a nonzero exit status.
• The existing read tool extracts page-marked PDF text, including in read-only review sessions. It needs no shell permission or external PDF executable. Use pages: "3" or pages: "3-5" to extract just the relevant pages of a long paper; offset and limit then count lines within that selection. Continuation notices retain the page range. Omitting pages preserves whole-document reading. Textless pages are explicitly marked; entirely textless, encrypted or malformed documents fail visibly. This is not OCR, figure inspection or layout validation. Selection skips text extraction outside the range, not file loading or document parsing, and does not validate unselected page content. No text cache is used.
• Notebook reads can use cells: "3" or cells: "1-3" for a source-first view of nbformat 4 notebooks. Start with cells: "1" to discover the cell count. The view includes saved execution counts and an output inventory, so long stored logs do not hide later validation code. Use includeOutputs: true for stored text/error outputs; rich MIME payloads are listed but not rendered. Saved outputs do not prove a fresh run or correctness. No code is executed or rewritten. Omit cells for the original raw JSON view, including exact editing context. offset/limit count rendered view lines; continuation notices retain the cell and output selection. The whole JSON file is still loaded and parsed; this is not a streaming JSON reader or a notebook execution engine.

The initial experiment changed only the default task prompt; PDF reading is the first subsequent tool capability; local pipeline status handling now also preserves execution failures. The Agent loop is unchanged. Small local prompt-profile trials showed reduced input usage, but also exposed timeouts, a provider request error, and missing persisted analysis scripts. These are not claims of general performance superiority. Improvements must be measured on matched tasks with failures and incomplete deliverables retained.

Build and run from source

Node.js 22.19+ and npm are required. This fork is currently a source preview: there is no separately published Argus-Pi npm package or binary release. Installing @earendil-works/pi-coding-agent from npm installs upstream Pi, not this fork.

git clone --branch argus https://github.com/Argus-AiTeam/Argus-Pi.git
cd Argus-Pi
npm ci --ignore-scripts
npm run hydrate:model-data
npm run build:offline
npm rebuild --workspace=@earendil-works/pi-coding-agent --ignore-scripts
./node_modules/.bin/argus-pi --help

For an existing Argus deployment configured to use the pi backend, place this checkout's node_modules/.bin first on that process's PATH. No Argus source change is required. Do not overwrite a working global Pi installation to try the fork. Both executable names retain Pi's existing configuration and authentication paths; use PI_CODING_AGENT_DIR when separate configuration is desired.

Use source control to update this source installation, not upstream's npm release or pi update --self. The CLI package is marked private until a dedicated publication identity and release process are established.

Ongoing development

The downstream default branch is argus; main initially retains the forked upstream history. Add https://github.com/earendil-works/pi.git as the upstream remote and merge reviewed upstream changes into argus in explicit updates. Do not force-reset the downstream branch to upstream.

Prioritize reproduced request failures and incomplete handoffs, then measured context, tool-loop and startup overhead. Preserve role isolation and stopping semantics. Each change needs a focused regression check and, for performance claims, same-model/same-budget task comparisons including failure counts. Do not commit credentials, private task logs, or provider authorization data.

CI also targets argus. Upstream contributor gates and release/catalog publication jobs are restricted to the upstream repository; this fork does not publish to upstream npm namespaces or infrastructure.

Upstream Pi

The original packages, documentation and MIT attribution are retained below.

The contributor approval policy below belongs to upstream Pi. Argus-Pi contributions target the argus branch; see CONTRIBUTING.md.

Pi Agent Harness

This is the home of the Pi agent harness project including our self extensible coding agent.

• @earendil-works/pi-coding-agent: Interactive coding agent CLI
• @earendil-works/pi-agent-core: Agent runtime with tool calling and state management
• @earendil-works/pi-ai: Unified multi-provider LLM API (OpenAI, Anthropic, Google, …)

To learn more about Pi:

• Visit pi.dev, the project website with demos
• Read the documentation, but you can also ask the agent to explain itself

All Packages

Package	Description
@earendil-works/chord	Standalone application-composition runtime for services, replicated state, RPC, and plugins
@earendil-works/pi-telemetry	Vendor-neutral telemetry contracts, reference adapter, conformance tests, and typed schemas
@earendil-works/pi-ai	Unified multi-provider LLM API (OpenAI, Anthropic, Google, etc.)
@earendil-works/pi-agent-core	Agent runtime with tool calling and state management
@earendil-works/pi-coding-agent	Interactive coding agent CLI
@earendil-works/pi-tui	Terminal UI library with differential rendering

For Slack/chat automation and workflows see earendil-works/pi-chat.

Permissions & Containerization

Pi does not include a built-in permission system for restricting filesystem, process, network, or credential access. By default, it runs with the permissions of the user and process that launched it.

If you need stronger boundaries, containerize or sandbox Pi. See packages/coding-agent/docs/containerization.md for three patterns:

• Gondolin extension: keep pi and provider auth on the host while routing built-in tools and ! commands into a local Linux micro-VM.
• Plain Docker: run the whole pi process in a local container for simple isolation.
• OpenShell: run the whole pi process in a policy-controlled sandbox.

Contributing

See CONTRIBUTING.md for contribution guidelines and AGENTS.md for project-specific rules (for both humans and agents). Longer term plans for Pi can also be found in RFCs.

Development

npm install --ignore-scripts  # Install all dependencies without running lifecycle scripts
npm run build         # Refresh model data, then build all packages
npm run build:offline # Rebuild using existing model data without network access
npm run check         # Lint, format, and type check
./test.sh            # Run tests (skips LLM-dependent tests without API keys)
./pi-test.sh         # Run pi from sources (can be run from any directory)

Building standalone binaries from release source

GitHub releases include a versioned source archive covered by the release's SHA256SUMS file. Extract it and run the same build script used for the official standalone binaries:

VERSION="<release-version>"
tar -xzf "pi-${VERSION}-source.tar.gz"
cd "pi-${VERSION}"
./scripts/build-binaries.sh --offline-model-data --platform linux-x64 --out "$PWD/out"

The archive includes release model data and native prebuilds. --offline-model-data uses that model data without refreshing provider catalogs. The script installs dependencies and builds the executable with its runtime assets; pass --skip-install if dependencies are already provided.

Supply-chain hardening

We treat npm dependency changes as reviewed code changes.

• Direct external dependencies are pinned to exact versions. Internal workspace packages remain version-ranged.
• .npmrc sets save-exact=true and min-release-age=2 to avoid same-day dependency releases during npm resolution.
• package-lock.json is the dependency ground truth. Pre-commit blocks accidental lockfile commits unless PI_ALLOW_LOCKFILE_CHANGE=1 is set.
• npm run check verifies pinned direct deps, native TypeScript import compatibility, and the generated coding-agent shrinkwrap.
• The published CLI package includes packages/coding-agent/npm-shrinkwrap.json, generated from the root lockfile, to pin transitive deps for npm users.
• Release smoke tests use npm run release:local to build, pack, and create isolated npm and Bun installs outside the repo before tagging a release.
• Local release installs, documented npm installs, and pi update --self use --ignore-scripts where supported.
• CI installs with npm ci --ignore-scripts, and a scheduled GitHub workflow runs npm audit --omit=dev plus npm audit signatures --omit=dev.
• Shrinkwrap generation has an explicit allowlist for dependency lifecycle scripts; new lifecycle-script deps fail checks until reviewed.

Share your OSS coding agent sessions

If you use Pi or other coding agents for open source work, please share your sessions.

Public OSS session data helps improve coding agents with real-world tasks, tool use, failures, and fixes instead of toy benchmarks.

For the full explanation, see this post on X.

To publish sessions, use badlogic/pi-share-hf. Read its README.md for setup instructions. All you need is a Hugging Face account, the Hugging Face CLI, and pi-share-hf.

You can also watch this video, where I show how I publish my pi-mono sessions.

I regularly publish my own pi-mono work sessions here:

• badlogicgames/pi-mono on Hugging Face

License

MIT

pi.dev domain graciously donated by


exe.dev