| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| manifest.json | 2026-08-05 | 1.7 kB | |
| timbal-2.4.0-linux-aarch64-gnu | 2026-08-05 | 1.0 MB | |
| timbal-2.4.0-linux-x86_64-gnu | 2026-08-05 | 1.1 MB | |
| timbal-2.4.0-macos-aarch64 | 2026-08-05 | 1.1 MB | |
| timbal-2.4.0-macos-x86_64 | 2026-08-05 | 1.2 MB | |
| timbal-2.4.0-windows-x86_64-gnu.exe | 2026-08-05 | 1.3 MB | |
| README.md | 2026-08-05 | 2.5 kB | |
| v2.4.0 source code.tar.gz | 2026-08-05 | 2.1 MB | |
| v2.4.0 source code.zip | 2026-08-05 | 2.6 MB | |
| Totals: 9 Items | 10.5 MB | 0 | |
What's Changed
- fix(types): drop assistant turns left empty after cross-provider replay (38af6be8)
- fix(core): let server-side tools coexist with client tools and survive provider switches (71025693)
- refactor(tests): move parallelism and files-integration tests to state/ and types/ (ea8b356d)
- feat(core): enable Anthropic automatic prompt caching by default — cache reads cost 0.1x (155242e4)
- refactor(tests): mirror package layout — move tracing/llm/collector tests out of tests/core (f2c31ed4)
- refactor(core): split llm_router into core/llm package — one module per concern, zero perf cost (3a9fb885)
- refactor(codegen,evals): extract shared helpers to kill copy-paste clones — net -187 lines (e763f511)
- fix(core): rename command-path loop variable shadowing the agent iteration counter (fbe517dc)
- fix(core): sync generators yielding None no longer truncate the stream (c994dc7a)
- perf(codegen): lazy transformer registration — CLI --help 381ms -> 60ms (2f7882ad)
- test: bump files-integration anthropic model to claude-sonnet-5 (38e4a710)
- perf(core): per-provider SDK isolation — anthropic-only runs never import openai (085695bb)
- perf(core): lazy provider-SDK imports — 'from timbal import Agent' 1.28s -> 0.39s (de7d4490)
- fix(server): don't pre-load voice ONNX models for non-voice deployments (b8e99481)
- fix(core): re-raise pending cancellation in agent tool fast paths (5923a4eb)
- bench: clear traces per latency iteration in long-conversation harness (9372eaaf)
- bench: refresh all cross-framework results after the perf branch (c922cb83)
- fix(core): restore caller's run context after nested top-level runnable (afbe96a0)
- feat(core): warn once when a sync handler blocks the event loop (4db236bb)
- perf(core): sequential fast path for linear workflows + shared _run_step (76db757e)
- perf(core): cache Message dumps + cut serialization overhead on long histories (8520ce19)
- bench: long conversations / long messages vs LangGraph (6d37bce1)
- refactor(core): decompose Runnable._stream into named lifecycle methods (dc6a52f2)
- perf(core): split call into raw _stream() + collector wrapper at the API boundary (ddab93c3)
- test(core): pin exact event-stream shapes for tools, agents, workflows (bbe5917e)
- perf(core): fast paths for non-streaming handlers, approvals-off, and single-tool dispatch (889852f6)
- perf(core): cut framework overhead 2-4x across agent and workflow hot paths (9d299cbb)
Full Changelog: https://github.com/timbal-ai/timbal/compare/v2.3.5...v2.4.0