Skip to content

Releases

HiveGate v0.4.1

Choose a tag to compare

@liberzon liberzon released this 15 Nov 21:01

Fixed

  • Gemini works again with google-genai 3.0, which pip install hivegate-ai==0.4.0 now
    resolves to. That release changed the client's default API version from v1beta to v1,
    where gemini-2.5 models refuse function calling (400) and preview models, including the
    default gemini-3-flash-preview, are not found (404). Every Gemini client HiveGate builds
    now pins v1beta: chat models, teams, the knowledge embedder, model listing and file
    uploads. Vertex AI (GOOGLE_GENAI_USE_VERTEXAI=true) keeps its own default, v1beta1.
    The 0.4.0 Docker image pins google-genai 2.34 and was not affected (#234)

Full changelog: v0.4.0...v0.4.1

HiveGate v0.4.0

Choose a tag to compare

@liberzon liberzon released this 11 Nov 18:09
987c40a

Upgrading

  • A model call that fails (a bad API key, an out-of-credits account, a provider outage) now
    returns HTTP 502 with status, error and detail from non-streaming agent chat and team
    runs. It used to return 200 with the provider's raw error as the answer. Streaming routes still
    return 200, and now end with an SSE error event carrying the same fields. A client that read
    the error text out of a 200 reply needs to change (#205, @rajshadow)
  • A chat agent configured with an MCP transport chat does not support (stdio, sse), or with
    an MCP URL that is not absolute HTTP or HTTPS, is now refused with HTTP 400. It used to run
    without those tools and say nothing. Supervisor and worker agents keep stdio (#206,
    @rayulumukku)

Fixed

  • POST /v2/agents/{agent_id}/session/clear now deletes the session's history. It used to return
    success and delete nothing (#161, @rajshadow)
  • docker compose up --wait no longer fails on the Qdrant healthcheck, which is removed
    (#154, @Arun100203)

Docs

  • DOCS_ENABLED and CORS_ORIGIN_LIST are documented (#145, @yusufthecoder474)
  • The Bright Data settings are documented (#147, @arpitkjaiswal)
  • The PyPI summary and README title name HiveGate as an AI agent runtime (#163)

Contributors

Full changelog: v0.3.1...v0.4.0

HiveGate v0.3.1

Choose a tag to compare

@liberzon liberzon released this 31 Oct 22:06
c00832f

First release on PyPI. Install with uv tool install hivegate-ai (or pipx install hivegate-ai). The package is hivegate-ai and the command is hivegate. This release also upgrades dependencies to clear the open security advisories (#134).

What's Changed

  • Fix the quickstart so it works from a fresh clone by @liberzon in #66
  • deps: update setuptools requirement from >=64 to >=84.0.0 by @dependabot[bot] in #60
  • deps: update langchain-core requirement from >=1.2.28 to >=1.6.5 by @dependabot[bot] in #61
  • deps: update langsmith requirement from >=0.7.30 to >=0.14.1 by @dependabot[bot] in #62
  • deps: update setuptools-scm requirement from >=10.0.5 to >=10.3.4 by @dependabot[bot] in #63
  • Add status badges, issue forms and contact links by @liberzon in #75
  • ci(dependabot): ignore websockets/portalocker majors capped by upstream by @liberzon in #58
  • docs(env): add DEFAULT_CHAT_MODEL to .env.example by @cestercian in #76
  • fix(deps): bump vulnerable packages in uv.lock by @liberzon in #79
  • deps: bump the python-minor group with 10 updates by @dependabot[bot] in #80
  • docs(settings): describe what set_cors_origin_list actually does by @rathi-mohit in #101
  • docs(readme): credit contributors by @liberzon in #89
  • test: ensure .env.example covers API settings by @shashank1027 in #103
  • docs(readme): credit @shashank1027 by @liberzon in #108
  • feat(anthropic): turn on prompt caching for Claude models by @liberzon in #110
  • feat(usage): meter the token cost of every model call by @liberzon in #109
  • docs(readme): document usage metering and prompt caching, add CodeTriage badge by @liberzon in #113
  • docs(readme): tag site links with utm_source by @liberzon in #115
  • fix(teams): cache paused team runs so they can be approved and committed by @liberzon in #116
  • test: drop a duplicated type: ignore that mypy rejects by @liberzon in #118
  • fix: honour AGENT_DEBUG_MODE for agent debug setting by @ompatil1906 in #114
  • deps: bump the python-minor group across 1 directory with 49 updates by @dependabot[bot] in #117
  • Show an out-of-credits provider error as a plain message, with error_code by @liberzon in #125
  • deps: upgrade everything to clear the security advisories by @liberzon in #134
  • fix: warn on unsupported MCP transports by @coughlmao in #146
  • feat: installable from PyPI with hivegate init and hivegate serve by @liberzon in #133

New Contributors

Full Changelog: v0.3.0...v0.3.1

HiveGate v0.3.0

Choose a tag to compare

@liberzon liberzon released this 29 Sep 19:21
473472c

First release of HiveGate, formerly Agents Gateway: an open-source FastAPI service that runs your Agno agents as a production API. 🌐 hivegate.dev

⚠️ Rename

The project, repo and image are now called HiveGate:

  • repo: hivegate-ai/hivegate
  • image: ghcr.io/hivegate-ai/hivegate
  • site: hivegate.dev

If you run an existing deployment:

  • set DB_DATABASE=agents_gateway to keep your current database
  • set OTEL_SERVICE_NAME=agents-gateway if your dashboards filter on the old name

See the CHANGELOG for details. (#59)

Highlights

  • Agno 3. Upgraded to agno 3.0.11 and mcp 2.2.0. MCP servers now connect through fastmcp. (#54, #55)
  • Native MCP tools on the chat path. (#12)
  • Models:
    • Claude 5 generation (#48)
    • GPT-6, Gemini 3.5–3.8, Grok, Z.ai GLM and DeepSeek (#57)
    • "latest" aliases per tier, resolved from each vendor's models API (#34)
    • a configurable default model, DEFAULT_CHAT_MODEL (#29)
  • Claude safety refusals are now surfaced instead of streaming an empty reply. (#49, #50)
  • GET /_/version returns the commit the running process was built from. (#56)

Fixes

  • DB connection pool exhaustion: every PostgresDb now shares one engine. (#37)
  • The SSE streamer no longer string-concatenates dict content. (#17)
  • The chat handler fetches prompts from whichever prompt backend is configured. (#4)
  • token_usage column names now match the ORM. (#5)

Maintenance

  • CI now tests the dependencies the image actually ships. (#53)
  • Dependency upgrades flagged by pip-audit. (#33)
  • Dependabot updates across Python packages and GitHub Actions.

Get started

git clone https://github.com/hivegate-ai/hivegate && cd hivegate
docker compose up -d
./scripts/dev_setup.sh && source .venv/bin/activate && ./scripts/start_server.sh

Or deploy ghcr.io/hivegate-ai/hivegate in one click to Render, Railway or Koyeb.