Backfill issue -144 — The AI Toolchain

EDITIONS tail grep head diff uniq

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.

VIEW

ISSUE VIEW full issue

Do you prefer this view?

$ tct list   # 2 tools matched

AI & LLM Tooling

◆ AI Agent Frameworks

Nous Research Hermes

v2026.3.28 NOTES

The agent that grows with you

Hermes v2026.3.28 adds Hugging Face inference, Telegram topic isolation, plugin lifecycle hooks, native Modal SDK, and a Nix flake.

└──▷ GET THIS VERSION

$ git clone --branch v2026.3.28 https://github.com/NousResearch/hermes-agent.git
# already have the repo? check out this version:
$ git checkout v2026.3.28

›Adds Hugging Face Inference API as a first-class provider with auth, setup wizard, curated agentic model picker, and live /models endpoint probe.
›Enables Telegram Private Chat Topics for project-based conversations with per-topic skill binding and isolated workflows within a single chat.
›Activates plugin lifecycle hooks — pre_llm_call, post_llm_call, on_session_start, and on_session_end — now firing in the agent loop and CLI/gateway.
›Replaces swe-rex dependency with native Modal SDK (Sandbox.create.aio + exec.aio), eliminating tunnel requirements for the Modal terminal backend.
›Adds GPT_TOOL_USE_GUIDANCE to enforce tool calls in GPT models instead of prose descriptions, and strips stale budget warnings from conversation history.

+15 moreshow less

›Introduces Nix flake with full uv2nix build, NixOS module with persistent container mode, and auto-generated config keys from Python source.
›Expands Nous Portal to 400+ available models via a single provider endpoint.
›Adds /resume CLI handler and reopen_session API for returning to previous sessions.
›Adds --source flag for third-party session isolation by origin.
›Introduces session search recent-sessions mode — omit query to browse recent sessions with titles, previews, and timestamps.
›Exposes compression.target_ratio, protect_last_n, and threshold in DEFAULT_CONFIG for ratio-based context compression tuning.
›Persists reasoning across gateway session turns with new schema v6 columns (reasoning, reasoning_details, codex_reasoning_items).
›Adds supply chain hardening: CI workflow scanning PRs for supply chain attack patterns, pinned dependency ranges, and regenerated uv.lock with hashes.
›Adds per-model native output limits for Anthropic models (128K for Opus 4.6, 64K for Sonnet 4.6) replacing hardcoded 16K max_tokens.
›Adds thinking-budget exhaustion detection to skip useless continuation retries when a model uses all output tokens on reasoning.
›Gives subagents independent iteration budgets.
›Surfaces all retry, fallback, and compression events to the user as formatted messages.
›Adds auto-discovery of fallback IPs via DNS-over-HTTPS when api.telegram.org is unreachable.
›Removes /model slash command from CLI and gateway in favor of hermes model subcommand.
›Increases API timeout default from 900s to 1800s to support slow-thinking models.

└──▷ BREAKING ON UPGRADE

!The /model slash command is removed from CLI and gateway; use the hermes model subcommand instead.
!The summary_target_tokens config key is replaced with ratio-based scaling; existing configurations referencing it will have no effect.
!The litellm dependency is removed; any external code or plugins that relied on it will break.

Was this useful?

◆ Local LLM Runtimes

llama.cpp

b8570 3 RELEASES · 2026-03-28 NOTES STABLE

llama.cpp b8570 adds reasoning_format = none support to the gpt-oss common layer.

└──▷ GET THIS VERSION

$ git clone --branch b8570 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8570

›Adds reasoning_format = none support to the gpt-oss common layer, allowing callers to suppress reasoning-format output entirely.

2 more releases in this issue · 2026-03-28

b8562 NOTES STABLE

llama.cpp b8562 adds /glob command to the CLI for batch file selection by pattern.

└──▷ GET THIS VERSION

$ git clone --branch b8562 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8562

›Adds /glob command to the CLI, enabling wildcard-based file selection including paths outside the current directory, with an error output when the maximum file limit is reached.

b8560 NOTES STABLE

llama-server gains --reuse-port flag to opt into SO_REUSEPORT for multi-process socket sharing

└──▷ GET THIS VERSION

$ git clone --branch b8560 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8560

└──▷ TRY IT

Run multiple llama-server instances on the same port so the OS load-balances incoming inference requests across processes.

$ llama-server --reuse-port -m model.gguf --port 8080

›Adds --reuse-port flag to llama-server, enabling SO_REUSEPORT on the listening socket to allow multiple server processes to bind the same port for load-balancing setups.

Was this useful?

// END OF TRANSMISSION

The AI Toolchain · curated daily · March 28, 2026
Releases & features from the open-source and commercial AI tools that practitioners actually run.

The AI Toolchain — issue -144, March 28, 2026

Nous Research Hermes

llama.cpp