idun-sdk
A thin, dependency-free Python client + CLI for Azure AI Foundry agents (Idun / NatureLM) and any OpenAI-compatible endpoint. Tenant-agnostic by design: bring your own Foundry resource — no hardcoded tenant coordinates.
Install
pip install idun-sdk
# with optional PDF ingest:
pip install "idun-sdk[pdf]" # pulls PyPDF2
Quick start (SDK)
from idun import IdunClient
client = IdunClient() # reads ~/.idun/config.toml
out = client.complete("Summarize the quarterly risk report.")
print(out["text"])
Two CLIs — two separate setup wizards (intentionally, not one)
This package ships two command-line tools with distinct purposes. They are
intentionally kept separate — idun is the Azure AI Foundry client,
idun-multi is the multi-provider LLM console. There is no unified
idun-wizard; each tool has its own first-run setup wizard:
idun— the Azure AI Foundry client. Agent completions, trajectory export, document matrix (idun matrix), prompt packs (idun run/idun packs), and Hugging Face Hub operations (idun hf). Azure-specific.idun wizardconfigures the Azure Foundry client (endpoint / project / agent) in~/.idun/config.toml.
idun-multi— the multi-provider LLM console. Talks to any of the 17 registered providers (OpenAI, Anthropic, Groq, OpenRouter, HF, …) plusrace,cost,models,doctor,support. Provider-agnostic.idun-multi wizardconfigures the default LLM provider (picks one of the 17 registered providers and prints the provider table so you can choose) in~/.idun/config.toml.
Both wizards write only to ~/.idun/config.toml (never ~/.idunrc,
which is no longer used). Credentials live in per-provider ~/.idun/<id>.token
files (0600).
idun-multi providers # list providers + credential status
idun-multi -p openrouter ask "hello" # -p goes BEFORE the subcommand
idun-multi race "explain quantum" # compare providers on one prompt
idun-multi doctor # environment + credential audit (also checks scripts)
idun-multi wizard # pick the default LLM provider
idun wizard # configure the Azure Foundry client
Async
import asyncio
from idun import AsyncIdunClient
async def main():
c = AsyncIdunClient()
out = await c.acomplete("What changed vs last quarter?")
print(out["text"])
asyncio.run(main())
CLI
idun — Azure AI Foundry client:
idun chat "your prompt" # one-shot completion
idun trace "your prompt" # full agent trajectory (steps)
idun export "your prompt" -o trace.md # run + save trajectory (json/md)
idun token # inspect / rotate stored Entra token
idun login # device-code login to your Foundry (azure/hf/openai)
idun status # show resolved backend + credential state
idun run <pack> <key> # run a bundled prompt pack
idun packs # list bundled prompt packs
idun matrix --docs DIR --questions FILE # Doc x Question pivot (IDEA α)
idun diff-docs --doc-a A --doc-b B --topics T # clause-drift compare (IDEA γ)
idun diff "A" "B" # compare two prompt trajectories side-by-side
idun hf whoami|status|push # Hugging Face Hub operations
idun openapi # print the bundled OpenAPI 3 spec for the completion API
idun logo # show bundled Foundry logo paths
idun welcome # ASCII banner
idun wizard # configure the Azure Foundry client
idun-multi — multi-provider LLM console (-p <provider> goes BEFORE the subcommand):
idun-multi providers # list providers + credential status
idun-multi -p openrouter ask "hello" # chat with a provider
idun-multi race "explain quantum" # compare providers on one prompt
idun-multi models # list models (or --discover)
idun-multi cost # token / cost estimate
idun-multi doctor # environment + credential audit
idun-multi verify # LIVE smoke-test configured providers
idun-multi wizard # pick the default LLM provider
idun-multi review <pr> [--post] [--inline] [--labels] [--no-cache]
# self-built PR reviewer (diff -> race -> artifacts)
Live provider verification (idun-multi verify)
The support matrix is honest about capability (which transport is wired),
but not about whether a provider has actually answered a request lately.
idun-multi verify performs a real, minimal API call against every provider
that has a credential configured, and records the outcome in
~/.idun/.verified.json (no secrets — only status, model, latency, and a
redacted error on failure). Unconfigured providers are reported as skip,
never fail, so an unconfigured machine shows an honest "not checked"
rather than a wall of false failures.
The recorded state feeds the Live column of idun-multi support and is
refreshed automatically every time you run race or verify. A provider with
Declared = — and Live = ? is simply unproven on this install — not
claimed broken, not claimed working.
idun-multi verify # all providers with a credential
idun-multi verify --providers openai,groq,anthropic
Self-built PR reviewer (idun-multi review)
A self-hosted alternative to CodeRabbit (ROADMAP Open #4). It fetches the PR
diff via gh, splits it into chunks, races them across up to 3 providers that
have credentials, and turns the results into actionable artifacts:
idun-multi review 123 # dry-run, prints findings
idun-multi review 123 --post # also post a PR comment
idun-multi review 123 --inline # post inline comments (GraphQL)
idun-multi review 123 --labels --post # add severity labels to the PR
idun-multi review 123 --no-cache # re-run all providers
Findings are parsed into severity/file/line and deduplicated across providers.
Chunk results are cached under ~/.idun/.review_cache.json (no secrets) so
repeat runs are cheap.
Supported providers
The idun CLI / IdunClient speak to any OpenAI-compatible endpoint plus the
native Azure AI Foundry transport. Registered providers (4 groups):
- OpenAI-compatible:
openai,groq,together,perplexity,fireworks,novita,xai,deepseek,mistral,openrouter,gemini,hf,nous - Anthropic:
anthropic(Claude) - Azure:
azure(Azure AI Foundry / Idun agent) - Local:
ollama,local(bring your own endpoint)
Each provider's capabilities (streaming, tools, vision, JSON mode) are derived
from the transports actually implemented in idun/providers.py — see
SUPPORT_MATRIX.md for the full, code-generated
matrix and per-transport capability details.
Hugging Face (hf): uses the OpenAI-compatible router
https://router.huggingface.co/v1 (the legacy api-inference.huggingface.co
host was retired). Requires an HF token (HF_TOKEN).
Any OpenAI-compatible base URL works with zero code changes — set it in the config and the provider switches automatically.
Configuration (tenant-agnostic)
~/.idun/config.toml (neutral defaults — supply your own resource). Secrets go
to per-provider ~/.idun/<id>.token files (mode 0600), not in this file:
# Written by `idun-multi wizard` (LLM provider default):
[defaults]
provider = "openrouter"
[openrouter]
model = "deepseek/deepseek-chat"
# Written by `idun wizard` (Azure AI Foundry client):
[azure]
base = "https://YOUR-RESOURCE.services.ai.azure.com"
agent = "your-agent-name"
project = "your-project"
No tenant coordinates are baked into the shipped code. The CLI reads the config
via idun.config; environment variables (IDUN_PROVIDER, OPENAI_API_KEY, …)
always win over the file.
Additional features
These build on the core SDK and live in
idun.matrix/ the playground repo.
Idun Matrix — Doc × Question pivot (IDEA α)
Build an N × M answer matrix over documents × questions. Each cell carries the answer, the source citation, and a status (GREEN = answered+cited, RED = contradiction, GRAY = no info).
idun matrix --docs ./contracts --questions ./questions.txt
Clause Drift compare (IDEA γ)
Compare two documents across topics and flag deviations:
idun diff-docs --doc-a contract_a.txt --doc-b contract_b.txt --topics topics.txt
PocketPal-style Bridge (IDEA β)
A tenant-agnostic mobile web UI that runs idun matrix against your Foundry
resource: https://qapdex-maker.github.io/idun-playground/matrix_app.html
Links
- Live demo (recorded trajectories + matrix UI): https://qapdex-maker.github.io/idun-playground/
- Matrix concept note: https://github.com/qapdex-maker/idun-playground/blob/main/DOC_MATRIX_CONCEPT.md
- Source: https://github.com/qapdex-maker/idun-sdk
- Changelog: https://github.com/qapdex-maker/idun-sdk/blob/main/CHANGELOG.md
Development
pip install -e ".[dev]"
pytest # offline; no network/API keys required
ruff check . # lint (pinned ruff==0.15.10 in CI)
License
MIT.
Metadata
Release files for idun-sdk 1.0.34
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| idun_sdk-1.0.34.tar.gz | 146.5 kB | Without provenance | Details for idun_sdk-1.0.34.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| idun_sdk-1.0.34-py3-none-any.whl | Python 3 | none | any | Details for idun_sdk-1.0.34-py3-none-any.whl |
Total release size: 252.7 kB
Release files / idun_sdk-1.0.34.tar.gz
| Download URL | idun_sdk-1.0.34.tar.gz |
|---|---|
| Size | 146.5 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
749cdaa5a7006f0eecd8e5b26fcaead01e04b01d48462b86b3f6b413bda97a07
|
|
BLAKE2b-256 checksum How to use checksums |
eaa73f81c311f2d0f62cfa4dba6bb4cc792804781d5863a09430658ae63aa2c3
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.14.6
|
Release files / idun_sdk-1.0.34-py3-none-any.whl
| Download URL | idun_sdk-1.0.34-py3-none-any.whl |
|---|---|
| Size | 106.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
871fb75543e74da5a021ab73345c4b0808bfc0b5a801c140ec88014ed15beb1e
|
|
BLAKE2b-256 checksum How to use checksums |
90d06c5745150c406681ca2f5875e56757b8dd45065716a7ab0cc69537e02ef4
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.14.6
|
Security for idun-sdk 1.0.34
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.
No provenance found
None of the files in this release were uploaded with provenance metadata.
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.