The AI Toolchain — issue -100, May 11, 2026

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 v2.1.139 adds agent view, /goal command, hook exec form, and subagent OTEL tracing.

$ git clone --branch v2.1.139 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:
$ git checkout v2.1.139

$ claude agents

$ /goal All unit tests pass and coverage is above 80%

• ›Adds claude agents command (Research Preview): a unified list of all Claude Code sessions — running, blocked, or complete.
• ›Adds /goal command: set a completion condition and Claude works autonomously across turns until it's met, with a live elapsed/turns/tokens overlay panel.
• ›Adds /scroll-speed command to tune mouse wheel scroll speed with a live preview.
• ›Adds claude plugin details <name> to inspect a plugin's component inventory and projected per-session token cost.
• ›Adds transcript view navigation shortcuts: ? for keyboard shortcuts, {/} to jump between user prompts, v to toggle the shortcut panel.

Glamourous agentic coding for all

Crush v0.67.0 adds a built-in shell interpreter for hooks, shell expansion in config values, and a touch tool for empty file creation.

$ git clone --branch v0.67.0 https://github.com/charmbracelet/crush.git
# already have the repo? check out this version:
$ git checkout v0.67.0

"api_key": "$(vault kv get my/secret/token)"

• ›Hooks now run via the built-in shell interpreter (mvdan/sh) by default, improving portability and Windows support; shebangs are still respected.
• ›Shell expansion now works in config values (e.g., "api_key": "$(vault kv get my/secret/token)") via the embedded interpreter, including MCP args, URLs, and LSP args/env.
• ›New write tool (via touch) lets the model create empty files, gated by a permission prompt for paths outside the working directory.

GitHub Copilot CLI v1.0.45 adds /autopilot and /fork slash commands, PowerShell fallback, and OpenTelemetry GenAI metrics.

$ git clone --branch v1.0.45 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v1.0.45

$ /autopilot

$ /fork

• ›Adds /autopilot slash command to toggle between interactive and autopilot modes mid-session.
• ›Adds /fork slash command to fork the current session into a new independent session.
• ›Falls back to powershell.exe (Windows PowerShell) when pwsh (PowerShell 7+) is not available on Windows.
• ›OpenTelemetry output now aligns with GenAI semantic conventions: MCP tool calls emit standard tool_call spans and a new gen_ai.client.operation.duration metric tracks tool execution time.
• ›CLI startup is up to ~1.5s faster on terminals with limited OSC color query support.

The open source coding agent.

OpenCode v1.14.47: Scout pre-materializes reference repos and large images are auto-resized before sending.

$ git clone --branch v1.14.47 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.14.47

• ›Scout can now materialize configured reference repositories up front so they are ready to search.
• ›Large image attachments are now auto-resized before sending, with configurable size limits.
• ›File and directory paths in the TUI now render relative to the session directory when possible.

LangChain Core 1.4.0 adds content-block streaming v2, ContextOverflowError, multimodal token counting, XML buffer formatting, and SSRF hardening.

$ git clone --branch langchain-core==1.4.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-core==1.4.0

async for chunk in model.astream_v2(messages):
    print(chunk)

from langchain_core.exceptions import ContextOverflowError

try:
    response = chat_model.invoke(long_messages)
except ContextOverflowError as e:
    print('Context limit exceeded:', e)

• ›Adds content-block-centric streaming API (stream_v2 / astream_v2), marked beta, for structured per-block streaming from chat models.
• ›Adds ContextOverflowError exception class, raised automatically by Anthropic and OpenAI integrations when context limits are exceeded.
• ›Adds multimodal support to count_tokens_approximately, enabling approximate token counting for image and other non-text content blocks.
• ›Adds tool-schema token counting to count_tokens_approximately, so tool definitions are included in approximate context estimates.
• ›Adds allow scaling by reported usage to count_tokens_approximately, letting callers calibrate estimates against actual usage metadata.

Semantic Kernel dotnet-1.76.0 adds ImageContent support in tool results and ExtraBody in OpenAI execution settings.

$ git clone --branch dotnet-1.76.0 https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:
$ git checkout dotnet-1.76.0

var settings = new OpenAIPromptExecutionSettings
{
    ExtraBody = new { reasoning_effort = "high", parallel_tool_calls = false }
};

• ›Adds ExtraBody property to OpenAIPromptExecutionSettings to pass arbitrary extra fields to the OpenAI API request body.
• ›Supports ImageContent in tool/function results, enabling connectors to return image data from tool calls.
• ›Adds deny-by-default AllowedUploadDirectories configuration to CloudDrivePlugin to restrict upload paths.

LocalAI v4.2.1 adds a DeepSeek V4 Flash backend with tool calls and KV cache, plus richer Ollama API model metadata.

$ git clone --branch v4.2.1 https://github.com/mudler/LocalAI.git
# already have the repo? check out this version:
$ git checkout v4.2.1

$ curl http://localhost:8080/api/show -d '{"name": "<model-name>"}'

• ›New ds4 backend adds DeepSeek V4 Flash support with tool calls, thinking mode, and KV cache.
• ›The /api/tags and /api/show Ollama-compatible endpoints now report model capabilities and details.

$ git clone --branch v4.2.0 https://github.com/mudler/LocalAI.git
# already have the repo? check out this version:
$ git checkout v4.2.0

$ OLLAMA_HOST=http://localhost:8080 ollama run qwen3

llama.cpp b9109 adds parallel speculative drafting with support for multiple chained speculators in the server and CLI.

$ git clone --branch b9109 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b9109

• ›Introduces common_speculative_type_from_names() to parse user-provided speculator types as a vector, enabling multiple speculative decoding strategies to be specified at once.
• ›Introduces common_get_enabled_speculative_impls() to determine which speculative decoding implementations are active based on the provided type vector.
• ›Introduces common_speculative_process() as a unified entry point for running multiple speculators sequentially, selecting the best draft by maximizing expected accepted tokens.
• ›Replaces the single type field in common_params_speculative with a vector of common_speculative_type values, allowing multiple speculator types to be configured in one pass.
• ›Adds parallel drafting support to the server, enabling concurrent speculative decoding across multiple request slots via a shared draft context.

Composio adds SHARED connected accounts with per-user ACL controls and a new updateAcl() method

$ 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]

composio.connectedAccounts.link(userId, authConfigId, {
  accountType: 'SHARED',
  aclConfigForShared: {
    allowAllUsers: false,
    allowedUserIds: ['user_alice', 'user_bob'],
    notAllowedUserIds: []
  }
});

composio.connectedAccounts.updateAcl('conn_nanoid_here', {
  notAllowedUserIds: ['user_eve']
});

• ›Adds accountType: 'SHARED' option to composio.connectedAccounts.link() — creates a shared connection usable by multiple userIds when explicitly pinned in a tool-router session; default remains 'PRIVATE'.
• ›Adds aclConfigForShared block ({ allowAllUsers, allowedUserIds, notAllowedUserIds }) on both create and retrieve for SHARED connections, with deny-win resolution: notAllowedUserIds checked first, then allowAllUsers, then allowedUserIds, otherwise deny-by-default.
• ›New composio.connectedAccounts.updateAcl(nanoid, { allowAllUsers, allowedUserIds, notAllowedUserIds }) method writes ACL via PATCH semantics — omit a field to leave it unchanged, pass an empty array to clear a list; at least one field required; each list accepts up to 1 000 entries, each userId up to 256 characters.
• ›Adds accountType and aclConfigForShared options to ToolRouterSession.authorize() so a SHARED connection with an ACL can be created in a single call from inside a tool-router session.
• ›The accountType field ('PRIVATE' | 'SHARED') is now returned in get() and list() responses for connected accounts.

Qdrant v1.18.0 adds TurboQuant 8x compression, named-vector CRUD APIs, low-memory mode, and a strict memory cap parameter.

$ git clone --branch v1.18.0 https://github.com/qdrant/qdrant.git
# already have the repo? check out this version:
$ git checkout v1.18.0

{
  "strict_mode_config": {
    "max_resident_memory_percent": 85
  }
}

$ curl -X PUT 'http://localhost:6333/collections/{collection_name}/vectors' \
  -H 'Content-Type: application/json' \
  -d '{
    "vectors": {
      "my-new-vector": {
        "size": 768,
        "distance": "Cosine"
      }
    }
  }'

• ›Adds max_resident_memory_percent strict-mode parameter to reject update requests when resident memory usage exceeds the configured threshold.
• ›Adds API endpoints to create and delete named vectors in an existing collection without recreating it.
• ›Adds config option to disable snapshot restore from URL, reducing the remote-fetch attack surface.
• ›Enforces API key / JWT authentication on internal gRPC endpoints.
• ›Introduces TurboQuant quantization variant delivering 8x vector compression with minimal recall degradation.

Weave v0.52.39 adds a GenAI observability schema with OTel span emission, agent-scoring events, and Session SDK ergonomics for manually-instrumented agents.

$ git clone --branch v0.52.39 https://github.com/wandb/weave.git
# already have the repo? check out this version:
$ git checkout v0.52.39

• ›Adds weave.score_agent_spans event emitted for GenAI turn_ended spans, enabling automatic scoring of agent turns in the observability pipeline.
• ›Adds TTL settings GET and POST API endpoints for managing trace time-to-live configuration.
• ›Adds agent span stats API endpoint for querying aggregated statistics over agent spans.
• ›Introduces a GenAI observability schema, extraction layer, and query layer for structured GenAI trace data.
• ›Wires OTel span emission into the Session SDK, expanding GenAI OTel coverage for agent sessions.