3 Releases published by 1 person

• 2025.8.0

  published Aug 6, 2025

• 2025.8.1

  published Aug 11, 2025

• 2025.8.2

  published Aug 12, 2025

32 Pull requests merged by 8 people

• fixed flaky e2e playwright test

  #3072 merged Aug 12, 2025

• Implement batched ClickHouse writes

  #2993 merged Aug 12, 2025

• bumped timeouts on docker compose for replicated CH

  #3066 merged Aug 12, 2025

• Allow users to edit variants in playground

  #2995 merged Aug 12, 2025

• bumped version to 2025.8.2

  #3064 merged Aug 12, 2025

• decreased log verbosity of migration information

  #3059 merged Aug 12, 2025

• Remove option type from TimeoutsConfig in ModelProvider

  #3034 merged Aug 12, 2025

• Allow specifying clickhouse url without user credentials

  #3055 merged Aug 12, 2025

• Fix ClickHouse cluster message

  #3057 merged Aug 11, 2025

• Recipe: Supervised Fine Tuning on Demonstrations with Unsloth

  #2273 merged Aug 11, 2025

• fixed hydration error and also tool parameters height issue

  #3056 merged Aug 11, 2025

• Allow dead code in optimizations tests

  #3054 merged Aug 11, 2025

• replaced simple cases of FINAL with LIMIT 1

  #3053 merged Aug 11, 2025

• Bump the rust-dependencies group with 9 updates

  #3051 merged Aug 11, 2025

• bumped version to 2025.8.1

  #3049 merged Aug 11, 2025

• Parse 'reasoning_content' field in VLLM responses

  #3018 merged Aug 11, 2025

• Update README.md with short demo videos

  #3024 merged Aug 9, 2025

• Use correct 'raw_response' when parsing Anthropic errors

  #3020 merged Aug 9, 2025

• fixed & tested rendering of function descriptions, also new rust erro…

  #3045 merged Aug 9, 2025

• Extract raw_response from AWS providers via TensorZeroInterceptor

  #3002 merged Aug 9, 2025

• Set ANTHROPIC_API_KEY when regenerating fixtures

  #3038 merged Aug 9, 2025

• added azure support for embeddings

  #3040 merged Aug 9, 2025

• bumped timeout in playground route to 30s

  #3041 merged Aug 9, 2025

• Start reporting 'tracing::error!' events to OTEL

  #3030 merged Aug 8, 2025

• add tensorzero.util reexport UUID

  #3017 merged Aug 8, 2025

• fixed tool calling in playground and added playwright test

  #3032 merged Aug 8, 2025

• add an OpenAI compatible embeddings endpoint

  #3022 merged Aug 8, 2025

• Add support for replicated ClickHouse setups

  #2895 merged Aug 7, 2025

• pulled feature flagging out of env file so we can use it on the client

  #2994 merged Aug 7, 2025

• Produce Thought/ThoughtChunk for Fireworks reasoning_content

  #3006 merged Aug 6, 2025

• use native client for inference in UI

  #2975 merged Aug 6, 2025

• Skip running migrations when TensorZeroMigration is up to date

  #3001 merged Aug 5, 2025

13 Pull requests opened by 7 people

• Add uuid7_str helper and improve uuid7 fallback

  #3019 opened Aug 7, 2025

• Improve the helm chart

  #3042 opened Aug 9, 2025

• set up automatic Helm publishing

  #3050 opened Aug 11, 2025

• feat(mock-inference-provider): add support for Together and GCP Verte…

  #3052 opened Aug 11, 2025

• Add OpenAI Reinforcement Fine-Tuning (RFT) Support

  #3060 opened Aug 11, 2025

• Add extra_body support for embedding models

  #3063 opened Aug 11, 2025

• Correctly handle os signals for ui container

  #3065 opened Aug 12, 2025

• fixed leftover comments referencing ContentBlockOutput

  #3068 opened Aug 12, 2025

• Add support for caching embeddings in OpenAPI compatible endpoint

  #3069 opened Aug 12, 2025

• consolidated all output components

  #3070 opened Aug 12, 2025

• gateway shutdown when CLICKHOUSE_URL is used

  #3071 opened Aug 12, 2025

• added default user to replicated

  #3074 opened Aug 12, 2025

• Store a Value directly in `StaticJSONSchema`

  #3075 opened Aug 12, 2025

21 Issues closed by 3 people

• Add the ability to edit prompt templates live in the UI Playground

  #2916 closed Aug 12, 2025

• Deprecate `JsonInferenceOutput` from `"~/utils/clickhouse/common`

  #2883 closed Aug 12, 2025

• TimeoutsConfig shouldn't be an option in ModelProviderConfig

  #3023 closed Aug 12, 2025

• Every JSON field (including tool parameters) in the inference detail page needs "Click to expand"

  #3021 closed Aug 11, 2025

• Fix Hydration errors in code editor component

  #2868 closed Aug 11, 2025

• 'Launch' button becomes unclickable in 'Launch Evaluations' modal

  #2909 closed Aug 11, 2025

• Add experimental/unstable support for exporting error messages over OTLP

  #3026 closed Aug 11, 2025

• Prefer `LIMIT 1` to `FINAL` in every query for tables without explicit versioning

  #2960 closed Aug 11, 2025

• Ensure UI SFT recipe can handle `include` directives in MiniJinja templates

  #1351 closed Aug 11, 2025

• Add an argument `"internal_dynamic_variant_config"` to the inference endpoint

  #2551 closed Aug 11, 2025

• Test ClickHouse migrations with large fixtures added before each migration

  #2104 closed Aug 11, 2025

• Expose the internal embeddings endpoint as directly callable by clients

  #2183 closed Aug 11, 2025

• Add support for `reasoning_content` field of vLLM responses

  #3004 closed Aug 11, 2025

• Add function descriptions to the UI

  #1988 closed Aug 9, 2025

• `tensorzero.util` should re-export `UUID` [Python SDK]

  #3015 closed Aug 9, 2025

• Gemini integrations need to resolve JSON schema $refs

  #3035 closed Aug 9, 2025

• Playground UI needs to handle tools correctly

  #3027 closed Aug 8, 2025

• Add support for reasoning summaries in GPT-5 series of models

  #3033 closed Aug 7, 2025

• Support automatic migrations for self-hosted replicated ClickHouse cluster

  #2228 closed Aug 7, 2025

• Add support for `reasoning_content` field of Fireworks responses

  #3005 closed Aug 6, 2025

• Start to skip migrations if `TensorZeroMigration` contains that migration.

  #2757 closed Aug 6, 2025

18 Issues opened by 3 people

• Allow users to configure `extra_headers` in embedding model providers in config

  #3076 opened Aug 12, 2025

• Create a better abstraction over our Python GIL and tokio interactions

  #3073 opened Aug 12, 2025

• Allow users to configure extra_body in embedding model providers in config

  #3062 opened Aug 11, 2025

• Implement variant writing to config from Rust

  #3061 opened Aug 11, 2025

• Integrate Helm deployments to Artifact Hub / R2 into CD

  #3048 opened Aug 11, 2025

• Add http mock tests for together and gemini SFT

  #3044 opened Aug 9, 2025

• Add load testing for clickhouse writes

  #3043 opened Aug 9, 2025

• Use a uv workspace for our `recipes` directory

  #3039 opened Aug 8, 2025

• Improve error handling of Rust inference interface.

  #3031 opened Aug 7, 2025

• Add `reasoning_effort` and `verbosity` to OpenAI model provider impl

  #3029 opened Aug 7, 2025

• Datapoints should keep track of static vs. dynamic tools

  #3028 opened Aug 7, 2025

• Support OTEL export to HTTP as well as gRPC

  #3025 opened Aug 7, 2025

• Improve error messages from Anthropic provider

  #3014 opened Aug 7, 2025

• Make every external type in the TensorZero Python SDK serializable

  #3011 opened Aug 6, 2025

• Dynamic evaluations example is broken

  #3010 opened Aug 6, 2025

• Add writes to TensorZeroMigration if a migration is skipped because `should_apply` is false

  #3009 opened Aug 6, 2025

• Force `skip_completed_migrations` to be true for TensorZero gateways.

  #3008 opened Aug 6, 2025

• Set default value for Claude 4.1 `max_tokens`

  #3003 opened Aug 6, 2025

30 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Add support for config globbing

  #2963 commented on Aug 12, 2025 • 8 new comments

• feat: Add native NVIDIA NIM provider integration (#767)

  #2954 commented on Aug 12, 2025 • 7 new comments

• Multimodal / Vision Fine-tuning Example

  #1827 commented on Aug 10, 2025 • 2 new comments

• Low Spec Demo

  #2974 commented on Aug 11, 2025 • 0 new comments

• Add GCPVertexGeminiSFTConfig to python OptimizationConfig type

  #2962 commented on Aug 12, 2025 • 0 new comments

• [Draft] OpenAI Agents SDK Integration

  #2933 commented on Aug 7, 2025 • 0 new comments

• [Issue-1961]: Handle clippy::match_same_arms

  #2800 commented on Aug 8, 2025 • 0 new comments

• Fully type-safe navigations

  #2797 commented on Aug 9, 2025 • 0 new comments

• Add variant filter

  #2724 commented on Aug 6, 2025 • 0 new comments

• Add exhausitve switch lint warning

  #2596 commented on Aug 9, 2025 • 0 new comments

• Bump the pip group across 21 directories with 5 updates

  #2502 commented on Aug 11, 2025 • 0 new comments

• Add a feature that allows JSON functions to retry if schema fails to validate

  #2835 commented on Aug 12, 2025 • 0 new comments

• Threshold field needs to handle "partial" numbers

  #2827 commented on Aug 12, 2025 • 0 new comments

• Validate responses from the OpenAI-compatible endpoint with the OpenAI OpenAPI spec

  #2387 commented on Aug 12, 2025 • 0 new comments

• Don't cache inferences if the output fails to parse

  #2924 commented on Aug 11, 2025 • 0 new comments

• Support structured multi-turn prompt templates and schemas

  #2972 commented on Aug 11, 2025 • 0 new comments

• Add `SETTINGS max_threads=1` to every point-lookup query that inspects a single granule

  #2959 commented on Aug 11, 2025 • 0 new comments

• Merge `NewOutput`, `NewOutputRust`, and `Output`

  #2885 commented on Aug 11, 2025 • 0 new comments

• Replace `ContentBlockOutput` with `ContentBlockChatOutput` in the UI

  #2882 commented on Aug 11, 2025 • 0 new comments

• `check-all-docker-compose.sh` should ensure health checks pass

  #2768 commented on Aug 11, 2025 • 0 new comments

• Migrate every optimization notebook to use `list_inferences` and `experimental_render_inferences`

  #2558 commented on Aug 11, 2025 • 0 new comments

• Handle unusual function/etc. names in the UI

  #1898 commented on Aug 11, 2025 • 0 new comments

• Update every Docker Compose example to ensure persistence of ClickHouse data

  #2421 commented on Aug 11, 2025 • 0 new comments

• Add a dedicated Ollama model provider

  #662 commented on Aug 9, 2025 • 0 new comments

• Local development and CI should use different database names for testing

  #1551 commented on Aug 9, 2025 • 0 new comments

• Add a Cohere model provider

  #655 commented on Aug 9, 2025 • 0 new comments

• `experimental_chain_of_thought` variant type doesn't support streaming

  #1839 commented on Aug 9, 2025 • 0 new comments

• Migration manager should not apply migrations if the database is already in a future state

  #2385 commented on Aug 8, 2025 • 0 new comments

• Implement a canonical function to convert rendered samples to external formats (e.g. ChatML)

  #2996 commented on Aug 7, 2025 • 0 new comments

• Default threshold in SFT form should be the median of the values for that

  #2904 commented on Aug 6, 2025 • 0 new comments