1.4.12 - apps drag'n'drop filestore, initial MCP support

What's Changed

• make frontend work better with filestore knowledge by @nessie993 in #668
• Model Context Protocol (MCP) support by @nessie993 in #680

security/refactoring/quality changes:

• Revive vol2 by @milosgajdos in #657
• (security): Bump golang.org/x/net module (CVE-2024-45338) by @milosgajdos in #658
• chore(deps): bump google.golang.org/grpc from 1.64.0 to 1.64.1 by @dependabot in #660
• (security): Bump golang.org/x/net module (CVE-2024-45338) by @milosgajdos in #661
• Bump gRPC module by @milosgajdos in #662
• (refactor): Client context plumbing by @milosgajdos in #663
• Add context plumbing for sessions. by @milosgajdos in #664
• Revive linter fixes by @milosgajdos in #665
• Enable revive vol4 by @milosgajdos in #666
• Rename vars so they follow idiomatic Go by @milosgajdos in #667
• Rename vars vol2 by @milosgajdos in #669
• Add comments to blank imports by @milosgajdos in #670
• Last revive fixes by @milosgajdos in #671
• Remove TOC from local-dev README by @milosgajdos in #673
• less confusing error for app api keys by @nessie993 in #674
• feat(test): initial smoke test by @philwinder in #675
• feat(test): slack notification by @philwinder in #676
• test(smoke): speed up and make tests more resilient by @philwinder in #677
• test(smoke): add installer test by @philwinder in #678

Full Changelog: 1.4.11...1.4.12

1.4.11 - out of the box local models support for gptscript; fix RAG over files on k8s

What's Changed

• Enable errcheck linter by @milosgajdos in #643
• fix: document serverUrl parameter in sample values.yaml by @lukemarsden in #648
• Add linter step into CI by @milosgajdos in #649
• Allow RUNNER_TOKEN to perform chat completions by @milosgajdos in #647
• Fix stack script by @milosgajdos in #650
• Update docker-compose file so gptscript uses RUNNER_TOKEN by @milosgajdos in #651
• fixes pdf parsing by @nessie993 in #652
• Remove redundant http.Flush when writing chunks by @milosgajdos in #653
• Make Go code Go code again: revive linter checks by @milosgajdos in #655
• chore(deps): bump transformers from 4.31.0 to 4.38.0 in /cog by @dependabot in #644
• refactor(scheduler): fix flakey test and refactor to make more testable by @philwinder in #656

Full Changelog: 1.4.10...1.4.11

1.4.10 - rag improvements

What's Changed

• Small error updates by @milosgajdos in #638
• chore(deps): bump golang.org/x/crypto from 0.27.0 to 0.31.0 in /operator by @dependabot in #639
• chore(deps): bump golang.org/x/crypto from 0.27.0 to 0.31.0 by @dependabot in #641
• add limits to the crawler by @nessie993 in #642
• Introduce golangci linter by @milosgajdos in #640
• fix rag search by @nessie993 in #645
• fix frontend code blocks by @nessie993 in #646

New Contributors

• @nessie993 made their first contribution in #642

Full Changelog: 1.4.9...1.4.10

1.4.9 - support function calling in helix runners, add gptscript runner

What's Changed

• Add gptscript_runner service to helix stack by @milosgajdos in #635
• Bump ollama and update LLM inference with Tool calling by @milosgajdos in #637

Full Changelog: 1.4.8...1.4.9

1.4.8 - Runner slimming, add llama3.3:70b

What's Changed

Runner slimming, add llama3.3:70b by @lukemarsden in #634:

• switch default model back to llama3.1:8b-instruct-q8_0, it was set to llama3.1:8b-instruct-q4_K_M by mistake
• remove rarely used models to reduce size of -large model, remove image model from -small, prune default startup models
• switch to symlinks instead of copying in the case that the model cache isn't bind-mounted (will dramatically reduce ephemeral storage usage in production)
• add new llama3.3:70b model which is apparently as good as llama3.1:405b
• upgrade to ollama v0.5.1
• add new OLLAMA_KV_CACHE_TYPE=q8_0 option to quantize kv store, should reduce long context memory usage (we can monitor memory usage in the dashboard and adjust the predicted memory values accordingly)

Full Changelog: 1.4.7...1.4.8

1.4.7 - [Breaking] Replace some Ollama Models to aid cache stability

Ollama regularly updates their "default" models to use new weights. For example llama3.1:7b gets updated roughly once every two weeks. Each time this happens we have to rebuild our base images from scratch.

To help avoid this, we've switched to use the same model, but with a more specific tag.

New sessions will use this new models by default. API users should change their model request to use the newer models.

Here is a list of changed models:

llama3.1:8b-instruct-q8_0
llama3.1:70b
aya:35b

Switching to:

llama3.1:8b-instruct-q4_K_M
llama3.1:70b-instruct-q4_K_M
aya:35b-23-q4_0

What's Changed

• refactor(runner): swap some ollama models to make them more stable and bust model cache to update by @philwinder in #631

Full Changelog: 1.4.6...1.4.7

1.4.6 - fix regression of frontend using chat models with togetherai

What's Changed

• Bump gptscript version by @milosgajdos in #624
• fix(scheduler): add background goroutine to check for dead runners by @philwinder in #623
• refactor(image): fix size of images by @philwinder in #626
• Use fork of gptscript and update prompt to user by @milosgajdos in #625

Full Changelog: 1.4.5...1.4.6

1.4.5 - Bugfix for openshift

What's Changed

• fix(diffusers): add --no-dev because it shouldn't install dev dependencies by @philwinder in #622

Full Changelog: 1.4.4...1.4.5

1.4.4 - New Image Generation Codebase and Capabilities

This release contains a new image generation codebase that allows Helix much more flexibility over what image generation models are added.

To celebrate this release we've baked a selection of the latest image generation models from Stability.AI and black-forest-labs into the -large runner images only.

In the future we will tinker with the baked model selection to reduce the size of the base images and add the code to allow fine-tuning.

What's Changed

• Small clean up of empty chars and unusued vars. by @milosgajdos in #614
• Remove llamaindex by @milosgajdos in #615
• chore(deps): bump github.com/hashicorp/go-retryablehttp from 0.7.4 to 0.7.7 in /operator by @dependabot in #616
• chore(deps): bump github.com/hashicorp/go-retryablehttp from 0.7.4 to 0.7.7 by @dependabot in #617
• chore: small cleanup by @milosgajdos in #618
• Update actionable system prompt by @milosgajdos in #619
• feat(runner): add basic diffusers server by @philwinder in #613

New Contributors

• @milosgajdos made their first contribution in #614
• @dependabot made their first contribution in #616

Full Changelog: 1.4.3...1.4.4

1.4.4 RC1 Diffusers Testing

1.4.4-rc1

chore(ci): bump runner base-images