Skip to main content

Jarvis CLI

Jarvis is a distributable terminal AI coding agent. Install it from PyPI with pipx or pip, or download a standalone executable from GitHub Releases. Cloning the repository is only required for development.

Efficient local-agent defaults

The local agent defaults to 20 steps, a 16,000-token input budget and a 6,000-token output budget. Adaptive multi-agent execution is off by default to avoid speculative extra model calls; enable it with JARVIS_ADAPTIVE_AGENTS=true or explicitly pass --multi-agent / JARVIS_MULTI_AGENT=true for tasks that benefit from parallel analysis. Override token ceilings with JARVIS_MAX_INPUT_TOKENS and JARVIS_MAX_OUTPUT_TOKENS when a task demonstrably needs more context. Adaptive mode also switches on the v0.7 structural task-enrichment stack, and JARVIS_ESCALATE_ON_FAILURE=true enables repair passes after a failed first attempt. On small local models keep both off: lean prompts reliably outperform the enriched multi-step stack.

Qwen3-family models always claim a reasoning capability, but reasoning mode on CPU burns thousands of tokens per turn before any answer appears. Jarvis therefore always requests think: false from the gateway unless JARVIS_THINK=true is set. Enable it only for tasks where slower, deliberate reasoning is worth the wall-clock cost.

Current contract line

Jarvis 0.11.4 consumes the provider-neutral Jarvis Core 0.17.3 contracts. All model requests use the direct inference API; the CLI has no AI Stack remote execution mode.

Jarvis Core 0.17.3 is published as an immutable PyPI release; the CLI pins that exact release so Jarvis and Core can evolve independently.

Jarvis is a standalone local coding agent. The agent loop and tools run on the user's machine, and all model requests go directly to the separately deployed jarvis-inference HTTP API. The Jarvis CLI does not call AI Stack and exposes no AI Stack remote execution mode. AI Stack is an independent sibling application that may also consume the same inference API.

Install

Jarvis is distributed as a normal Python application. You do not need to clone the Git repository to use it.

Python 3.10+ is required. pipx installs Jarvis into an isolated environment and exposes the jarvis command globally.

python3 -m pip install --user pipx
python3 -m pipx ensurepath
pipx install jarvis-agent-cli
jarvis --version

Upgrade later with:

pipx upgrade jarvis-agent-cli

Standard pip

python3 -m pip install jarvis-agent-cli
jarvis --version

Standalone executable

Download the platform-specific jarvis-* executable from GitHub Releases. Releases provide Linux amd64, macOS arm64 and Windows amd64 builds, plus SHA256 checksums and build provenance.

On Linux/macOS:

chmod +x ./jarvis-*
./jarvis-* --version

On Windows, run the .exe directly from PowerShell or Command Prompt.

Development install

Only contributors working on the source tree need a repository checkout:

git clone https://github.com/abdullahalrifat/jarvis.git
cd jarvis
python3 -m pip install -e .

Jarvis 0.11.4 consumes jarvis-agent-core==0.17.3 from PyPI. The dependency is pinned to the exact Core release in pyproject.toml.

See docs/install.md for the complete distribution and upgrade guide.

Token-efficient runtime

Core 0.17.3 provides provider-neutral primitives for bounded context construction, token/cost estimation, route budgets, adaptive routing and empirical route calibration. Jarvis keeps inference execution policy in the CLI, while shared efficiency and calibration contracts remain provider-neutral in Core.

Context construction can prioritize required task state, recent tool evidence and relevant files under an explicit budget. Route decisions can account for estimated token cost, latency, risk and measured task outcomes.

Real workload efficiency benchmark

Jarvis owns local runtime observations, task-level workload definitions and quality evaluation. AI Stack owns telemetry and calibration for server-side Runs. Both adapt their own measured outcomes to the provider-neutral jarvis-core observation/calibration contract. The benchmark covers CI triage, provider architecture, release readiness, PR review, documentation alignment and efficiency audits.

Use the corpus to compare local-only and automatic routes using success, quality, incorrect completions, tool failures, latency, input/output/cache tokens and estimated cost. Runtime evidence should only influence automatic routing after the Core minimum-sample and quality-floor safeguards are satisfied.

Connect directly to inference

Configure Jarvis to use the dedicated inference gateway. No AI Stack, database, vector store, or web UI is required for local repository work.

export INFERENCE_BASE_URL=http://<inference-vm-ip>:8080/v1
export INFERENCE_API_KEY=<your-inference-secret>
export JARVIS_MODEL=qwen3:1.7b

jarvis model-doctor
cd /path/to/repository
jarvis "review this repository and fix the highest-impact issue"

Bare tasks use the standalone local agent. The agent loop, repository tools, permission checks and verification run on the machine where Jarvis is installed; only model requests go to the inference gateway.

AI Stack boundary

Jarvis does not expose AI Stack remote commands and never requires AI_STACK_BASE_URL or AI_STACK_API_KEY. Use Jarvis for local repository work; deploy AI Stack independently when you need its own durable Runs API, queues, retrieval, integrations or UI. Both applications call jarvis-inference directly. See docs/architecture.md for ownership boundaries.

Application boundaries

Need Use
Local repository work, interactive terminal, local automation Jarvis
Durable shared Runs, queues and integrations AI Stack independently

Jarvis and AI Stack are independent consumers of jarvis-agent-core and jarvis-inference; neither calls the other.

Implemented runtime

The runtime includes bounded repository/Git tools, allowlisted command execution, sandbox/network policy, sessions/checkpoints, proof records, multimodal context, hierarchical instructions/memory, Skills/Hooks, deny-by-default MCP, repository/LSP intelligence, worktrees, plan mode, adaptive context, failure memory, multi-agent execution, heterogeneous routing, browser verification, background jobs/cron and observability.

Development

python -m pip install -e . -r requirements.txt
black --check --diff src tests
ruff check src tests --select E9,F63,F7,F82
pytest -q --cov=. --cov-report=term-missing --cov-fail-under=70
python -m build

Additional guides:

Production request path

Normal bare-task usage is Jarvis CLI -> jarvis-inference -> model backend. AI Stack is an optional consumer of the same inference contract.

Metadata

Release files for jarvis-agent-cli 0.12.3

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for jarvis-agent-cli 0.12.3
File Size Uploaded Provenance Actions
jarvis_agent_cli-0.12.3.tar.gz 174.7 kB With provenance Details for jarvis_agent_cli-0.12.3.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for jarvis-agent-cli 0.12.3
File Interpreter ABI Platform Actions
jarvis_agent_cli-0.12.3-py3-none-any.whl Python 3 none any Details for jarvis_agent_cli-0.12.3-py3-none-any.whl

Total release size: 343.4 kB

Release files / jarvis_agent_cli-0.12.3.tar.gz

Download URL jarvis_agent_cli-0.12.3.tar.gz
Size 174.7 kB
Tags Source
SHA-256 checksum
How to use checksums
bd0d03b089e3298420e8045cf464adca57ed0411bc3fdc0ad8a1b9bc59d47100
BLAKE2b-256 checksum
How to use checksums
bd0fb5bb08f8da20b51584e9ce783fab5e60ee692daf48e25512c650a07ed975
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 8, 2026.

Transparency log

Release files / jarvis_agent_cli-0.12.3-py3-none-any.whl

Download URL jarvis_agent_cli-0.12.3-py3-none-any.whl
Size 168.8 kB
Tags Python 3
SHA-256 checksum
How to use checksums
ecd68880b7574aedbad0eabf1f719770eb1751b73a6b5f559c5c0a2021b02e88
BLAKE2b-256 checksum
How to use checksums
587ce2f0cf9d6b46277fdb0eeb7ebc090fc4adcbc2d2e4f2fe64555ae103dd82
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 8, 2026.

Transparency log

Security for jarvis-agent-cli 0.12.3

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for jarvis_agent_cli-0.12.3.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 8, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page