typesafeainate/dspy-typesafeify
User repository on GitHub
Add a decorator for dspy Signatures that automatically uses TypeSafe where relevant
- Forks
- 3
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Repos & SDKs
typesafeify (DSPy decorator)
A DSPy fork whose @typesafeify decorator routes decision-shaped signature fields to Jev, so bool, Literal, and scored outputs come back typed while freeform fields still use the LM.
- Stars
- 66
- Language
- Python
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- typesafeainate
- Last verified
- Sep 2026
About this project
This is a deliberately stripped-down proof-of-concept fork of DSPy. It exists to test one idea in isolation: can a normal DSPy signature opt into Typesafe's typed inference path with one decorator, while application code keeps constructing and calling dspy.Predict exactly as before? The meaningful integration diff is intentionally tiny: The caller does not learn a new predictor API: @typesafeify(...) reads the signature's output annotations and builds a hybrid execution plan.
Quoted from the project's README. Read the full README.
Works forJev for Scoring
Used inJev for Software & DevTools
Highlights
- One decorator turns bool, Literal, and configured score fields into a single Jev request, while string fields stay with DSPy's LM.
- Across three observed tickets the decorated run averaged 1.958s versus 2.329s for plain DSPy, 15.9% faster.
- Modeled cost fell from $0.000377 to $0.000263 per ticket, a 30.1% reduction.
- The demo checks field and instruction parity between the baseline and decorated signatures before either request runs.
Quickstart
python
from typesafe_dspy import typesafeify
@typesafeify(score_fields={"severity_score": [1, 5]})
class SupportTicketTriage(dspy.Signature):
ticket: dict = dspy.InputField()
customer_impacting: bool = dspy.OutputField()
owner_team: Literal["api-platform", "billing"] = dspy.OutputField()
severity_score: float = dspy.OutputField()Watch out
MIT-licensed, but a deliberately stripped-down proof-of-concept fork rather than a replacement for upstream DSPy; needs both OpenAI and TypeSafe keys.
Questions about this entry
Short answers drawn from the linked project or page.
No. The README calls it a deliberately stripped-down proof-of-concept fork and points you back to the upstream project and dspy.ai for real work. It exists to test one idea in isolation, whether a normal DSPy signature can opt into a typed inference path with a single decorator.
Partly modeled. The timings come from running the same three tickets through both paths, but the cost table prices Typesafe input tokens at $0.042 per million as a supplied estimate to be replaced with a public pricing source, and Typesafe output tokens are reported but not assigned a cost. Both paths also need an OpenAI key.
The demo keeps separate baseline and decorated signature files precisely so a controlled before-and-after is possible, and before either request it proves the two signatures have identical fields and instructions. The result is a hybrid path: string fields are still generated by the LM, after the typed results are already known.
Only the decision-shaped fields move. bool outputs become Noul probabilities thresholded into booleans, Literal outputs become Choice decisions with per-option probabilities, and a score field you configure becomes a Score mapped back onto its declared range. Freeform strings stay with DSPy's LM, and callers still build dspy.Predict as before.
Run uv sync --extra typesafe, set OPENAI_API_KEY and TYPESAFE_API_KEY, then OPENAI_MODEL=gpt-5.6-luna uv run python examples/typesafe_dspy_ticket_triage/run_demo.py --limit 3 --show-document; it prints the execution plan, signature parity, typed results, timings, tokens, and cost per case.
bool becomes a Noul thresholded into a boolean, Literal becomes a Choice with per-option probabilities, and a field you name in score_fields (for example severity_score: [1, 5]) becomes a Score mapped back onto that range.
No. Application code keeps constructing and calling dspy.Predict exactly as before; the decorator builds a hybrid execution plan, and freeform string fields are still generated by the DSPy LM after the typed results are known.
More Jev resources like this
A filesystem-first framework for durable AI agents whose auto model router defaults to Jev through Vercel AI Gateway and whose evaluate helper asks typed Choice, Score, and Boolean questions inside tools.
Jev Repos & SDKs#community#typescript#agents
Community
vercel/eveTypeScript
vercel/eveTypeScriptA local server and Python package that runs open models with Hugging Face Transformers or PyTorch and returns Choice, Score, and Noul answers read from next-token logits instead of generated JSON.
Jev Repos & SDKs#community#python#open-models
Community
featherless-ai/simple-jevPython
featherless-ai/simple-jevPythonAn Apache-licensed System One decision server that reads Choice, Score, and Noul probabilities from DiffusionGemma 26B-A4B through vLLM on NVIDIA or MLX on Apple silicon, so TypeSafe SDKs work unchanged.
Jev Repos & SDKs#community#python#open-models
Community
razorback16/openjevPython
razorback16/openjevPythonA Juspay TypeScript SDK that treats Jev as a third inference type beside generate and stream: providers declare which kinds they serve, and tryDecide returns typed boolean, choice, or score answers, null when no decision provider is set.
Jev Repos & SDKs#community#typescript#sdk
Community
juspay/neurolinkTypeScript
juspay/neurolinkTypeScriptA Home Assistant integration that turns Jev questions about entity state into sensors and automation actions, with an Assist conversation agent and a daily token budget.
Jev Repos & SDKs#community#python#home-assistant
Community
AboveColin/HA-JevPython
AboveColin/HA-JevPythonA Python library that turns Jev-style decision models into ordinary function calls: check returns a bool, classify picks one option, label picks every option that applies, and score rates on a scale, with probabilities behind every answer.
Jev Repos & SDKs#community#github#python
Community
jlowin/vibecheckPython
jlowin/vibecheckPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.