Open sourcePython · ≈39k stars · MIT
DSPy
Programming, not prompting: optimize the whole pipeline.
DSPy introduced programming models for LLM pipelines — declarative modules whose prompts and weights are optimized against metrics. Its optimizer-first philosophy reshaped how production harnesses treat prompts.
Pros & cons
Each point cites its sources and says what kind of evidence backs it. A second reviewer tried to refute every claim before it was kept.
Pros
- User reportsPractitioners say they hear consistently good things from DSPy users, and that companies without it often end up building a half-baked version themselves.[1][2]
- User reportsDeclarative signatures avoid upfront abstraction cost and encourage good program structure.[1][2]
- MetadataMIT licensed, a Stanford NLP research pedigree, and steady releases (3.4.1 on 2026-10-12).[1]
Cons
- User reportsNeeds an automated metric and train/eval datasets upfront, which is hard for agents and chat.[1][2]
- User reportsOptimised prompts can feel like opaque blobs, and the compile step hurts reproducibility when you swap models.[1][2]
- User reportsDynamically typed signatures are unergonomic in typed codebases, and the built-in agent loop is weak.[1]
- User reportsConcurrency bugs: dspy.Parallel breaks streaming, and ParallelEvaluate fails on long evals.[1][2]
- Best for
- Pipelines with a measurable metric where you want systematic prompt/program optimisation.
- Watch out for
- Without evals and datasets it adds little; it is not a full agent harness.
Assessed Oct 29, 2026 · adversarially reviewedSuggest a correction
Activity
- Last push
- · 2d ago
- Latest release
- 3.5.0
· 11d ago - Last 90 days
- 4 releases
- · 11d ago
DSPy 3.5: A Lighter, Faster Foundation DSPy 3.5 brings a lighter LM layer, graduated core modules, and new History and image capabilities for RLM. Upgrading from 3.4.1 This release changes prompts, defaults and APIs. Read the upgrade guide for breaking changes, migration steps…
- · 5w ago
DSPy 3.4.1 DSPy 3.4.1 makes GEPA's structured InstructionProposer the default instruction proposer and lets dspy.RLM build DSPy sub-agents from inside its REPL. It also adds Python 3.15 support, fixes a 3.4.0 import crash, and includes streaming, adapter, and interpreter fixes.…
- · 8w ago
DSPy 3.4.0 DSPy 3.4 introduces Jev integration through TypeSafe, with three experimental decision types—Noul, Choice, and Score—that bring probability evidence into ordinary DSPy signatures. ReAnchor calibrates those decisions against your program's metric. This release also…
3.4.0b1Pre-release
· 2mo agoDSPy 3.4.0 Beta 1 DSPy 3.4.0b1 is the first beta of 3.4. It moves language-model execution to a shared engine interface, adds a local CPython interpreter for trusted code, and brings async execution to ReActV2. It also adds custom Flex code proposals in GEPA and fixes…
- · 3mo ago
DSPy 3.3.1 DSPy 3.3.1 contains many interpreter fixes and improvements. It makes PythonInterpreter easier to install, substantially strengthens sandbox isolation and request handling, and adds end-to-end visibility into interpreter execution. The release also improves optimizer…
Checked against GitHub on Nov 21, 2026. Release notes are shortened to plain text.
In research (8)
- Agent-Native by Design: A Programming Language for Agentic ApplicationsVerdict: Incremental
Nov 16, 2026 · Baseline framework compared on LOC, builder tokens and task quality.
- GEPA-Curator: Curated Learning for Self Improving AgentsVerdict: Useful
Oct 27, 2026 · GEPA, the optimizer this paper extends, ships as a DSPy optimizer, so the Curator is a natural add-on for DSPy users.
Oct 27, 2026 · The benchmarks used (HotpotQA, HoVer, IFBench) follow DSPy/GEPA program-optimisation setups, so EvalAlloc applies directly to DSPy-style optimisers.
- Hierarchical Skill Evolution for Coding AgentVerdict: Useful
Oct 26, 2026 · GEPA, the reflective prompt-evolution baseline HSE beats, comes from the DSPy ecosystem.
- RSIV-Harness: A Recursive Self-Improvement Harness Framework for Open-Domain Conversational Text-to-VisualizationVerdict: Incremental
Oct 26, 2026 · Programmatic prompt/pipeline optimisation, the main alternative to agent-written harness revisions.
- AT-AT: Automatic Tool Optimization from Agent TracesVerdict: Useful
Oct 26, 2026 · GEPA, an optimiser from the DSPy ecosystem, is a main baseline.
Showing the 6 newest of 8 papers.
Neighbours on the workbench
TypeScript · ≈395k stars · MIT
openclaw/openclaw
The viral open-source personal AI assistant that really does things.
Maintained by OpenClaw Foundation
v2026.11.6releasedTypeScript · ≈285k stars · MIT
deepseek-ai/deepseek-harness
Everything is a Plugin — DeepSeek's open agent harness.
Maintained by DeepSeek
dsh-v0.2.2-rc.5 prereleasedTypeScript · ≈212k stars · Sustainable Use License (fair-code; GitHub reports NOASSERTION)
n8n-io/n8n
Fair-code workflow automation platform with native AI capabilities.
Maintained by n8n-io
n8n@1.123.100releasedConnections
Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.
Maintained by · 1
- Stanford NLPOrganisation
Mentioned in · 1
- AI Engineer Sydney 2026Event · 2026-12-07
