The daily firehose — everything the toolchain shipped today, already filtered.
// HOW THIS ISSUE IS MADE
We read every release from the 174 tools on our watchlist at the source — GitHub and GitLab release notes, vendor release pages and changelogs, project blogs and feeds, vendor press releases, and the source code behind the tag. Bug-fix-only releases and non-product newsroom noise are dropped; what's left is summarized down to the new capability, how to try it, and any screenshots or videos the release itself published. Every entry links to the sources it was built from.
Lightweight coding agent that runs in your terminal
Codex CLI gains Windows proxy-only egress enforcement, piped stdin for codex exec, dynamic bearer tokens for custom model providers, and device-code ChatGPT login.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.118.0 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.118.0
└──▷ TRY IT
Pipe a file's contents into Codex while supplying a distinct prompt — useful for scripted code review or transformation tasks in CI.
$ cat main.rs | codex exec "Refactor this to use idiomatic error handling"
›Enforces proxy-only networking on Windows sandboxes with OS-level egress rules, replacing environment-variable-only controls.
›Adds ChatGPT device-code sign-in flow for app-server clients when browser-callback login is unavailable.
›Supports prompt-plus-stdin piping in codex exec, letting you pipe input while passing a separate prompt on the command line.
›Enables custom model providers to fetch and refresh short-lived bearer tokens dynamically instead of relying on static config or environment credentials.
OpenCode v1.3.11 adds a dedicated Kimi model system prompt and hardened plugin install safety.
└──▷ GET THIS VERSION
$ git clone --branch v1.3.11 https://github.com/sst/opencode.git
# already have the repo? check out this version:$ git checkout v1.3.11
›Adds a dedicated system prompt for Kimi models, improving response quality for that model family.
›Pins explicit plugin versions on install and blocks package install scripts, reducing supply-chain risk in extensions.
›Skips plugins that lack a matching server or TUI entrypoint with a warning instead of a hard failure, and applies default options from package exports on install.
Letta 0.16.7 raises the default context window to 128k, adds conversation forking, new model support, and a recompile endpoint.
└──▷ GET THIS VERSION
$ git clone --branch 0.16.7 https://github.com/letta-ai/letta.git
# already have the repo? check out this version:$ git checkout 0.16.7
└──▷ TRY IT
Force a recompile of a conversation's system prompt after editing a memory block via the API, so the agent sees the updated context immediately.
$ curl -X POST https://<your-letta-server>/v1/conversations/<conversation-id>/recompile
›Adds POST /v1/conversations/{id}/recompile endpoint to trigger system-prompt recompilation after manual block edits.
›Adds conversation forking — fork any conversation, including the default, with shared message history.
›Adds is_byok flag to error responses for easier debugging of Bring-Your-Own-Key failures.
›Raises the global context window default from 32k to 128k for self-hosted servers with unknown models.
›Adds sort-by-last_message_at for conversation listings.
+12 moreshow less
›Adds OTID-based idempotent conversation streaming for retry safety.
›Adds per-request system prompt overrides via request-scoped system overrides.
›Adds Baseten as a provider with full frontend integration, serverless auto mode, and reasoning support.
›Adds full support for GPT-5.4, including mini, nano, and fast variants.
›Adds GLM-5, GLM-5.1, GLM-5 Turbo, and GLM-4.7 model support.
›Adds MiniMax M2.7 model support.
›Adds Fireworks and zAI coding provider support.
›Adds projection-style rendering for git memory (memfs) content in system prompts.
›Blocks internal MCP server targets.
›Adds WebSocket transport for the OpenAI Responses API.
›Adds readiness enforcement scaffold with request pressure, DB pool, SSE lifecycle, and event loop lag monitoring (M1-M3 metrics pipeline).
›Moves multi-agent tools to a less-privileged execution environment.
└──▷ BREAKING ON UPGRADE
!Block limit validation has been removed from the git memory sync path — blocks can now grow freely, and any per-turn cost caps you enforced via block limits must be managed by other means.
›Adds QQ Bot as a bundled channel plugin with multi-account setup, SecretRef-aware credentials, slash commands, reminders, and media send/receive support.
›Adds native Slack approval routing so exec approval prompts can be authorized directly in Slack instead of falling back to the Web UI or terminal.
›Adds openclaw flows list|show|cancel commands for linear task flow control, surfacing orchestrated work with doctor recovery hints for orphaned flows.
›Unifies ACP, subagent, cron, and background CLI execution under one SQLite-backed task ledger with audit, maintenance, and status visibility.
›Adds remote HTTP/SSE server support for mcp.servers URL configs, including auth headers and safe credential redaction.
+15 moreshow less
›Adds draft streaming for Matrix so partial replies update the same message in place rather than sending a new message per chunk.
›Adds explicit channels.matrix.proxy config for routing Matrix traffic through an HTTP(S) proxy, including account-level overrides.
›Adds optional channels.matrix.historyLimit for Matrix group triggers, with per-agent watermarks and retry-safe snapshots.
›Adds LINE image, video, and audio outbound sends on the LINE-specific delivery path, including preview/tracking handling for videos.
›Adds WhatsApp emoji reactions so agents can react to incoming messages instead of sending a text reply.
›Adds a Graph-backed member info action for Microsoft Teams to resolve channel member details directly from Microsoft Graph.
›Adds per-agent memorySearch.qmd.extraCollections so agents can opt into cross-agent session search without collapsing all transcripts into one namespace.
›Adds native Codex web search support for embedded Pi runs, including config, docs, and wizard coverage.
›Adds Android notification-forwarding controls with package filtering, quiet hours, and rate limiting.
›Adds per-agent threadReplies overrides for Matrix DMs to keep thread session isolation aligned with the triggering message policy.
›Adds bundle MCP tools with provider-safe names (serverName__toolName), optional streamable-http transport selection, and per-server connection timeouts.
›Adds a configurable idle-stream timeout for embedded LLM runner requests so stalled model streams abort cleanly.
›Adds structured provider diagnostics and fallback attempt analytics for TTS.
›Adds an explicit default-off ACPX plugin-tools MCP bridge config with documented trust boundary.
›Adds OpenAI text.verbosity forwarding across Responses HTTP and WebSocket transports, surfaced in /status.
└──▷ BREAKING ON UPGRADE
!The nodes.run shell wrapper is removed from the CLI and agent nodes tool; node shell execution must now go through exec host=node.
!Legacy provider compat subpaths, older bundled provider setup, and channel-runtime compatibility shims in the Plugin SDK are deprecated and emit migration warnings; use openclaw/plugin-sdk/* entrypoints and local api.ts/runtime-api.ts barrels going forward.
!Built-in dangerous-code critical findings and install-time scan failures now fail closed by default; plugin installs and gateway-backed skill dependency installs that previously succeeded may now require --dangerously-force-unsafe-install to proceed.
!trusted-proxy in Gateway/auth now rejects mixed shared-token configs, and local-direct fallback requires the configured token instead of implicitly authenticating same-host callers.
!Node commands now stay disabled until node pairing is approved; device pairing alone no longer exposes declared node commands.
!Node-originated runs now stay on a reduced trusted surface; notification-driven or node-triggered flows that previously relied on broader host/session tool access may need adjustment.
PydanticAI v1.74.0 adds online evaluation infrastructure, TextContent metadata, time-sortable run IDs, and MCP Server instructions support.
└──▷ GET THIS VERSION
$ git clone --branch v1.74.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:$ git checkout v1.74.0
└──▷ USE IT
Attach invisible metadata to a user prompt — useful for tagging messages with session or request context without leaking it to the model.
python
from pydantic_ai.messages import TextContent
prompt = TextContent(text='Summarize this document.', metadata={'session_id': 'abc123', 'user_tier': 'pro'})
result = await agent.run([prompt])
›Adds AbstractToolset.get_instructions method and include_instructions argument to MCP Servers, enabling toolsets to surface dynamic instructions to agents.
›Adds TextContent class for user prompts, supporting a metadata field that is attached to the content object but not sent to the model.
›Introduces online evaluation infrastructure for pydantic-evals, enabling live/online evaluation workflows.
›Makes agent run IDs time-sortable and propagates agent name and run ID as span attributes on all agent run child spans, improving observability.
LanceDB python-v0.30.2 adds a type-safe expression builder API, progress bars for add(), parallel remote inserts, and Float16/Float64/Uint8 vector query support.
└──▷ GET THIS VERSION
$ git clone --branch python-v0.30.2 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:$ git checkout python-v0.30.2
›Adds a type-safe expression builder API for Python (feat(python): add type-safe expression builder API).
›Adds a progress bar for the add() method when ingesting data.
›Enables parallel inserts for remote tables via multipart write (Rust backend).
›Supports Float16, Float64, and Uint8 vector queries in the Node.js client.
LanceDB v0.27.2 adds parallel remote inserts, Float16/Float64/Uint8 vector queries, a type-safe Python expression builder, and a progress bar for add().
└──▷ GET THIS VERSION
$ git clone --branch v0.27.2 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:$ git checkout v0.27.2
›Adds type-safe expression builder API in Python for constructing queries with compile-time safety.
›Supports Float16, Float64, and Uint8 vector queries in the Node.js client.
›Adds progress bar for add() operations to surface ingestion status.
›Enables parallel inserts for remote tables via multipart write in the Rust client, improving throughput for large uploads.
›Adds session_id field to the scores blob storage export pipeline.
›Adds dataset_run_id field to the scores blob storage export pipeline.
›Extends LANGFUSE_DELETE_SKIP_PROJECT_IDS (renamed from LANGFUSE_TRACE_DELETE_SKIP_PROJECT_IDS) to also apply to score deletions, not just trace deletions.
›Adds boolean score type support for LLM-as-a-judge evaluators, alongside the existing numeric and categorical types.
›Allows LLM-as-a-judge evaluators to filter observations by tool names and tool call count.
+6 moreshow less
›Adds Redis command and socket timeouts via new configuration for fail-fast behavior during Redis outages.
›Adds SSE-based query progress streaming for dashboard widgets in the frontend.
›Adds a 'Run Experiment' dialog directly on the Experiments page.
›Adds a duplicate folder action for prompts.
›Adds a universal $4K default spend alert for all cloud plans.
›Applies trace-equivalent rate limits to score deletion API endpoints.
└──▷ BREAKING ON UPGRADE
!The environment variable LANGFUSE_TRACE_DELETE_SKIP_PROJECT_IDS is renamed to LANGFUSE_DELETE_SKIP_PROJECT_IDS; self-hosted instances using the old name must update their configuration.
THE AI TOOLCHAIN