paxlet

AI Cost Tracking

• 🤖 LLM usage: $0.0714 (3 commits)
• 👤 Human dev: ~$190 (1.9h @ $100/h, 30min dedup)

Generated on 2026-09-22 using openrouter/qwen/qwen3-coder-next

Build small. Connect everything.

Paxlet Core is a deliberately small contract for packaging resources and actions as addressable capabilities.

A Paxlet has a stable identity, a version, clear inputs/outputs, explicit dependencies and requested permissions. A node can resolve it, verify it, run it and produce a receipt.

URN identity
    │
    ▼
┌──────────── Paxlet ────────────┐
│ resources                     │
│ actions                       │
│ dependencies                  │
│ permissions                   │
│ provenance                    │
└───────────────────────────────┘
    │
    ▼
URI locations → Node → Receipt

Start in five minutes

Requires Python 3.11+ and no third-party runtime dependencies. You can run it directly after cloning:

./bin/paxlet init /tmp/my-first
./bin/paxlet run /tmp/my-first hello --input '{"name":"Ada"}'

./bin/paxlet verify examples/hello
./bin/paxlet run examples/hello hello --input '{"name":"Ada"}'

Optional developer install:

python -m venv .venv
. .venv/bin/activate
pip install -e .

Or address the same capability by identity:

./bin/paxlet resolve urn:paxlet:example:hello
./bin/paxlet run urn:paxlet:example:hello hello --input '{"name":"Ada"}'

Try a science-oriented example:

./bin/paxlet run examples/science-fasta-gc analyze --input '{}'
./bin/paxlet run examples/statistics-mean mean --input '{"values":[1,2,4,8]}'

CLI

paxlet init    DIRECTORY
paxlet verify  PATH|URN
paxlet inspect PATH|URN
paxlet resolve URN
paxlet run     PATH|URN ACTION --input JSON
paxlet pack    PATH|URN -o package.paxlet.zip

Why Core is small

Paxlet is not an AI-agent framework. It is the portable unit that larger systems can use.

Paxlet Core      package + identity + contract
Paxlet Node      resolution + grants + execution + receipts
Paxlet Network   registries + peers + distribution
Taskand          planning + orchestration + adaptive-system showcase

Taskand can demonstrate planners, processes, federation, mesh, Digital Twin, MCP, governance and recovery without making those features mandatory for every Paxlet implementation.

Repository map

SPEC.md                 normative Core 0.1 draft
schemas/                manifest + receipt schemas
paxlet/                 tiny Python reference runtime
examples/               five understandable capabilities
registry/local.json     local URN → URI resolver example
conformance/            implementation compatibility runner
docs/                   addressing, security, receipts, layers
integrations/taskand/    how Taskand fits as the showcase

Conformance

python -m unittest discover -s tests -v
python conformance/run.py

Important security note

The reference runtime is intentionally educational. It validates package paths and exposes secrets only after explicit user grant, but it does not enforce filesystem/network permissions with an OS sandbox. Production nodes should use containers, VMs or OS policy for enforcement. See docs/security.md.

Status

Core 0.1 is a draft intended to make the idea testable and implementable by independent runtimes. See ROADMAP.md.

License

Licensed under Apache-2.0.