The AI Toolchain — issue -118, April 23, 2026

Agno v2.6.0 adds HITL for Teams and Workflows, runtime Factories, multi-framework AgentOS support, and a new Context Provider API.

$ git clone --branch v2.6.0 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:
$ git checkout v2.6.0

$ curl -X GET 'https://<agentOS-host>/sessions?type=workflow' \
  -H 'Authorization: Bearer <token>'

• ›Adds AgentFactory, TeamFactory, and WorkflowFactory for dynamically creating Agents, Teams, and Workflows at runtime, enabling multi-tenant use cases.
• ›Adds agno.context — a first-party API for plugging external sources (filesystem, web, SQL database, Slack, Google Drive, MCP server) into an agent as a natural-language tool.
• ›Adds an API layer for Team human-in-the-loop (HITL) with support in the AgentOS chat page, including Team Approvals.
• ›Adds executor-level HITL support for Workflow steps (WorkflowExecutor) when a pause-tool flow is configured on an agent or team within a workflow step.
• ›Adds reconnection and resume capability for Agent/Team runs using SSE in AgentOS, allowing interrupted sessions to continue from where they left off.

The agent that grows with you

Hermes v2026.4.23 adds a full Ink TUI, native AWS Bedrock, 5 new inference providers, QQBot, and a vastly expanded plugin surface.

$ git clone --branch v2026.4.23 https://github.com/NousResearch/hermes-agent.git
# already have the repo? check out this version:
$ git checkout v2026.4.23

$ hermes --tui

$ /steer focus only on files under src/auth/ and skip any database migrations

• ›Adds hermes --tui: a full React/Ink rewrite of the interactive CLI with sticky composer, live streaming, OSC-52 clipboard support, per-turn stopwatch, git branch status bar, /clear confirm, light-theme preset, and subagent spawn observability overlay.
• ›Adds pluggable agent/transports/ layer with AnthropicTransport, ChatCompletionsTransport, ResponsesApiTransport, and BedrockTransport — each owning its own format conversion and API shape.
• ›Adds native AWS Bedrock provider support via the Converse API.
• ›Adds five new inference providers: NVIDIA NIM, Arcee AI, Step Plan, Google Gemini CLI OAuth, and Vercel ai-gateway (with pricing and dynamic discovery).
• ›Adds GPT-5.5 access via ChatGPT Codex OAuth, with live model discovery wired into the model picker so new OpenAI releases appear without catalog updates.

langchain-fireworks 1.2.0 adds streaming usage metadata, a standard model property, and new model-profile fields.

$ git clone --branch langchain-fireworks==1.2.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-fireworks==1.2.0

• ›Populates usage_metadata on streaming responses, enabling token-count tracking during streamed Fireworks calls.
• ›Adds a standard model property to the Fireworks chat model class, aligning it with other LangChain partner integrations.
• ›Adds text_inputs and text_outputs fields to model profiles, exposing explicit modality metadata per model.
• ›Honors max_retries on Fireworks LLM/chat model instances, making retry configuration effective.

$ git clone --branch langchain-core==1.3.1 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-core==1.3.1

OpenAI Agents SDK v0.14.5 adds an idle timeout option for Modal sandbox environments.

$ git clone --branch v0.14.5 https://github.com/openai/openai-agents-python.git
# already have the repo? check out this version:
$ git checkout v0.14.5

• ›Adds an idle timeout option for Modal sandbox code execution environments, allowing control over how long a sandbox remains active without activity.

Your own personal AI assistant. Any OS.

OpenClaw v2026.4.22 adds xAI multimodal support, new STT providers, /models add command, WhatsApp reply quoting, and a local embedded TUI mode.

$ git clone --branch v2026.4.22 https://github.com/openclaw/openclaw.git
# already have the repo? check out this version:
$ git checkout v2026.4.22

$ /models add xai grok-imagine-image-pro

• ›Adds xAI image generation (grok-imagine-image, grok-imagine-image-pro), reference-image edits, text-to-speech (six voices, MP3/WAV/PCM/G.711), and grok-stt audio transcription including realtime Voice Call streaming.
• ›Adds Voice Call streaming transcription for Deepgram, ElevenLabs, and Mistral; ElevenLabs also gains Scribe v2 batch audio transcription for inbound media.
• ›New local embedded TUI mode lets you run terminal chats without a Gateway while keeping plugin approval gates enforced.
• ›Onboarding now auto-installs missing provider and channel plugins so first-run setup completes without manual plugin recovery.
• ›OpenAI Responses models automatically use OpenAI's native web_search tool when web search is enabled and no managed search provider is pinned.

PydanticAI v1.86.0 adds UIAdapter.manage_system_prompt and ReinjectSystemPrompt for dynamic system prompt control in UI adapters.

$ git clone --branch v1.86.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:
$ git checkout v1.86.0

• ›Adds UIAdapter.manage_system_prompt method and ReinjectSystemPrompt capability, enabling UI adapters to control and reinject system prompts at runtime.

SRE Agent - CNCF Sandbox Project

HolmesGPT 0.25.0 adds a Crossplane toolset and OAuth support for MCP servers in CLI and server mode.

$ git clone --branch 0.25.0 https://github.com/HolmesGPT/holmesgpt.git
# already have the repo? check out this version:
$ git checkout 0.25.0

• ›Adds OAuth support for MCP servers in both CLI and server mode.
• ›Adds a Crossplane toolset for investigating Crossplane resources.

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

Claude Code v2.1.119 adds GitLab/Bitbucket PR support, persistent config, richer hooks, and OpenTelemetry improvements.

$ git clone --branch v2.1.119 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:
$ git checkout v2.1.119

$ claude --from-pr https://gitlab.com/myorg/myrepo/-/merge_requests/42

$ CLAUDE_CODE_HIDE_CWD=1 claude

• ›Adds prUrlTemplate setting to point the footer PR badge at a custom code-review URL instead of github.com.
• ›Adds CLAUDE_CODE_HIDE_CWD environment variable to hide the working directory in the startup logo.
• ›--from-pr now accepts GitLab merge-request, Bitbucket pull-request, and GitHub Enterprise PR URLs in addition to github.com.
• ›/config settings (theme, editor mode, verbose, etc.) now persist to ~/.claude/settings.json and participate in project/local/policy override precedence.
• ›--print mode now honors the agent's tools: and disallowedTools: frontmatter, matching interactive-mode behavior.

$ git clone --branch v2.1.118 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:
$ git checkout v2.1.118

$ DISABLE_UPDATES=1 claude

{
  "type": "mcp_tool",
  "server": "my-mcp-server",
  "tool": "log_event",
  "params": { "event": "agent_stop" }
}

$ claude plugin tag v1.2.0

GitHub Copilot CLI v1.0.35 adds HTTP hooks, named sessions, new slash commands, and MCP/LSP improvements.

$ git clone --branch v1.0.35 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v1.0.35

$ gh copilot --name my-feature-work
# later...
gh copilot --resume=my-feature-work

continueOnAutoMode: true

hooks:
  - event: tool_call
    http:
      url: https://hooks.example.com/copilot
      method: POST

• ›Adds COPILOT_GH_HOST environment variable to set the GitHub hostname, taking precedence over GH_HOST.
• ›Adds --name flag to name sessions and --resume=<name> to resume them by name.
• ›Adds continueOnAutoMode config option to automatically switch to auto model on rate limit instead of pausing.
• ›Adds /session delete, /session delete <id>, and /session delete-all subcommands, plus x-to-delete in the session picker.
• ›Adds HTTP hook support, allowing hooks to POST JSON payloads to a configured URL instead of running a local command.

an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM

Goose v1.32.0 adds Exa AI search, two new AI providers, @agent mentions, /skills and /edit commands, voice dictation, and Unix socket MCP transport.

$ git clone --branch v1.32.0 https://github.com/block/goose.git
# already have the repo? check out this version:
$ git checkout v1.32.0

• ›Adds Exa AI-powered web search as a built-in tool.
• ›New /skills command to browse and manage available agent skills from the CLI.
• ›New /edit command for on-demand prompt editing mid-session.
• ›Supports @agent mentions in chat to target specific agents directly.
• ›Adds Novita AI as a declarative provider.

Lightweight coding agent that runs in your terminal

Codex CLI v0.124.0 adds Amazon Bedrock support, stable hooks, quick TUI reasoning controls, and remote plugin marketplace access.

$ git clone --branch rust-v0.124.0 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.124.0

$ # Inside the Codex TUI, press Alt+, to lower reasoning or Alt+. to raise it

[hooks]
[[hooks.post_tool_use]]
tool = "apply_patch"
command = "ruff check --fix ."

• ›Adds Alt+, / Alt+. keyboard shortcuts in the TUI to lower or raise reasoning level on the fly; model upgrades now reset reasoning to the new model's default.
• ›Adds first-class Amazon Bedrock support for OpenAI-compatible providers, including AWS SigV4 signing and AWS credential-based authentication.
• ›Hooks are now stable: configurable inline in config.toml and requirements.toml, and can observe MCP tools, apply_patch edits, and long-running Bash sessions.
• ›App-server sessions can now manage multiple environments and select an environment and working directory per turn, enabling precise multi-workspace and remote targeting.
• ›Remote plugin marketplaces can now be listed and read directly, with larger result pages and more reliable detail lookups.

$ git clone --branch rust-v0.123.0 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.123.0

$ /mcp verbose

The open source coding agent.

OpenCode v1.14.22 lets projects store a custom icon override that persists across sessions.

$ git clone --branch v1.14.22 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.14.22

• ›Supports .npmrc settings during npm installs, enabling private registry and auth configs to be honoured automatically.
• ›Enables per-project custom icon overrides that persist correctly across sessions.

$ git clone --branch v1.14.21 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.14.21

AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI

Pi v0.70.0 adds GPT-5.5 Codex support, fuzzy auth-provider search, auth source labels, and granular built-in tool disabling.

$ git clone --branch v0.70.0 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:
$ git checkout v0.70.0

$ pi --no-builtin-tools

• ›Adds openai-codex/gpt-5.5 as a model option with xhigh reasoning support and priority-tier pricing.
• ›Adds fuzzy search/filtering to the /login provider selector, making it faster to find providers when many are configured.
• ›Adds auth source labels in /login showing whether credentials come from --api-key, an environment variable, or a custom provider fallback — without exposing secrets.
• ›Adds --no-builtin-tools CLI flag and createAgentSession({ noTools: "builtin" }) SDK option to disable only built-in tools while keeping extension tools active.
• ›Adds opt-in OSC 9;4 terminal progress indicators, toggled via terminal.showTerminalProgress in /settings.

Qwen Code v0.15.1 combines elapsed time and timeout into a single shell time indicator.

$ git clone --branch v0.15.1 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.15.1

• ›Combines elapsed time and timeout into a unified shell time indicator in the CLI, giving at-a-glance visibility into how long a command has run versus its allowed limit.

An open-source AI agent that brings the power of Gemini directly into your terminal.

gemini-cli v0.39.0 adds /memory inbox skill review, invoke_subagent unification, Plan Mode confirmation, and scroll momentum enhancements.

$ git clone --branch v0.39.0 https://github.com/google-gemini/gemini-cli.git
# already have the repo? check out this version:
$ git checkout v0.39.0

$ /memory inbox

{
  "mcpServers": {
    "my-server": {
      "url": "https://my-mcp-server.example.com",
      "auth": {
        "type": "bearer",
        "token": "<your-token>"
      }
    }
  }
}

• ›Adds /memory inbox command for reviewing and patching extracted skills, including skill patching support with inbox integration.
• ›Introduces unified invoke_subagent tool, consolidating subagent invocation under a single interface.
• ›Requires user confirmation for activate_skill actions in Plan Mode, adding a safety gate before skill activation.
• ›Implements silent fallback for Plan Mode model routing when the preferred model is unavailable.
• ›Adds useAgentStream hook and wires it into AppContainer for streaming agent responses.

Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

Ollama v0.21.2 bundles web search in OpenClaw and stabilizes the ollama launch onboarding flow.

$ git clone --branch v0.21.2 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.21.2

$ ollama launch

• ›OpenClaw integration now includes Ollama's web search plugin out of the box, enabling web-augmented queries without manual plugin setup.

llama.cpp b8902 enables the transcriptions API for LFM2-Audio models in the server.

$ git clone --branch b8902 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8902

• ›Enables the transcriptions API for LFM2-Audio models in the server.

$ git clone --branch b8893 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8893

oobabooga textgen v4.6.2 adds tool-call confirmation buttons, stdio MCP server support, and a --preserve-thinking flag.

$ git clone --branch v4.6.2 https://github.com/oobabooga/textgen.git
# already have the repo? check out this version:
$ git checkout v4.6.2

$ python server.py --preserve-thinking

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/home/user/projects"]
    }
  }
}

• ›Adds --preserve-thinking CLI flag and matching chat template parameter preserve_thinking to control whether thinking blocks from prior turns are kept in context.
• ›Adds stdio MCP server support via user_data/mcp.json, using the same format as Claude Desktop and Cursor, enabling local subprocess-based MCP servers alongside existing HTTP MCP servers.
• ›Adds inline approve/reject/always-approve tool-call confirmation buttons, enabled via the new 'Confirm tool calls' checkbox in the Chat tab.
• ›Adds UI checkbox for preserve_thinking to control thinking-block retention without the CLI.
• ›Passes --draft-min 48 by default for llama.cpp draftless speculative decoding.

Composio core 0.6.11 adds file-upload path hardening with denylist controls and a pre-upload hook.

$ git clone --branch @composio/[email protected] https://github.com/ComposioHQ/composio.git
# already have the repo? check out this version:
$ git checkout @composio/[email protected]

const tools = await composio.tools.get({
  sensitiveFileUploadProtection: true,
  fileUploadPathDenySegments: [".secrets", "vault"],
  beforeFileUpload: (filePath) => {
    if (filePath.includes("/tmp/staging")) return filePath.replace("/tmp/staging", "/tmp/safe");
    return filePath;
  }
});

const tools = await composio.tools.get({
  sensitiveFileUploadProtection: false
});

• ›Adds automatic blocking of local file uploads from credential-sensitive paths (e.g. .ssh, .aws) and credential-like filenames (e.g. .env, default SSH private-key names) by default; URLs and File objects are unaffected.
• ›Adds sensitiveFileUploadProtection: false option to opt out of the default credential-path blocking when required.
• ›Adds fileUploadPathDenySegments option to extend the built-in denylist with custom path segments.
• ›Adds an optional beforeFileUpload hook (e.g. on composio.tools.get) to rewrite paths, return false to abort, or throw to cancel an upload programmatically.
• ›Introduces two new error types: ComposioSensitiveFilePathBlockedError (path matched the denylist) and ComposioFileUploadAbortedError (hook returned false or threw).

$ git clone --branch [email protected] https://github.com/ComposioHQ/composio.git
# already have the repo? check out this version:
$ git checkout [email protected]

from composio import Composio
from composio.modifiers import before_file_upload

@before_file_upload
def audit_upload(path: str) -> str | bool:
    if 'confidential' in path:
        return False  # aborts upload, raises FileUploadAbortedError
    print(f'Uploading: {path}')
    return path

client = Composio(
    sensitive_file_upload_protection=True,
    file_upload_path_deny_segments=['.ssh', '.aws', '.env'],
)
tools = client.tools.get(actions=[...], modifiers=[audit_upload])

Langfuse v3.170.0 adds unstable evals public API endpoints, a DELETE endpoint for LLM connections, configurable OIDC token signing, and new blob storage export frequencies.

$ git clone --branch v3.170.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.170.0

$ curl -X DELETE https://<your-langfuse-host>/api/public/llm-connections/<connection-id> \
  -H 'Authorization: Bearer <secret-key>'

• ›Adds unstable evals public API endpoints for triggering and managing evaluations programmatically.
• ›Adds DELETE endpoint for LLM connections via the public API.
• ›Adds DATABASE_URL special-character warning on migration failure to surface encoding issues early.
• ›Supports configuration of the ID Token signed response algorithm for OIDC/auth via new config option.
• ›Emits .rate and .time metrics with shard tags for DataDog aggregation, enabling per-shard observability.

Weaviate v1.37.2 speeds up collection export snapshots and adds asymmetric distance computation to hfresh.

$ git clone --branch v1.37.2 https://github.com/weaviate/weaviate.git
# already have the repo? check out this version:
$ git checkout v1.37.2

• ›Speeds up collection export snapshotting via concurrent tenant de-activation, reducing snapshot time for multi-tenant collections.
• ›Adds asymmetric distance computation to the hfresh index, improving approximate nearest-neighbor search accuracy for quantized vectors.