Daily · star velocity · 2026-09-24

Decision models keep the lead, with agent memory, an agent control plane and a graph database close behind

Today's list: the two decision models from yesterday still lead, hindsight brings learning agent memory, nasiko fronts every A2A agent with one Rust process, and hydradb puts a graph database on object storage.

Fastest-growing this period

Counts come from GitHub's official stargazers/history endpoint: the stars gained on each day of the window. Project briefs and the reading notes below are editorial additions.

1

NandhaKishorM/laya : Answer typed questions about any text in one 33-millisecond forward pass
Python
A non-autoregressive decision engine that reads text, an email or a JSON document and returns a typed answer, a choice, a score or a yes/no flag, in one 33-millisecond forward pass. Three checkpoints cover English, over one hundred languages and typed workflows, and a built-in router picks one per request. Nothing is generated, so there is nothing to parse or hallucinate.

+3,480
2

jaredpalmer/kev : Small open decision models answering choice, yes/no and scoring questions in one request
Python
A family of small decision models built on Qwen3.5, with 0.8B, 4B and 9B variants you can run or train yourself. One request shares a single input across choice, yes/no and rating questions, each answered with probabilities instead of a single label. It serves locally on CUDA, ROCm and Apple Silicon, and a web playground shows how option order changes the answers.

+1,499
3

vectorize-io/hindsight : Agent memory with three operations: retain, recall, and reflect to keep learning
Python
An agent memory system organized around three operations: retain observations, recall them when needed, and reflect to keep what an agent knows up to date. Memory banks and mental models structure the knowledge, it runs as a Docker, pip or embedded Python service against 25+ LLM providers, and an MCP server plus a two-line LLM wrapper connect it to coding agents.

+1,450
4

google/ax : Declare agent tasks in YAML, sandbox them, then SSH into a running one
Go
A declarative orchestrator from Google for agent workloads on a cluster, in the spirit of Kubernetes. One YAML file declares the task, pre-wired Git repos and MCP servers, a network allowlist and the model; ax apply launches it into a sandbox with CPU and memory limits, and ax ssh opens a shell inside a running agent to watch what it does.

+1,420
5

hydra-db/hydradb : A Rust graph database that keeps the graph on S3 and scales compute separately
Rust
A distributed graph database written in Rust where S3-compatible object storage is the durable source of truth and compute splits into query-serving data nodes plus background indexers that rebuild from durable state. OpenCypher queries run against pinned snapshots with GraphBLAS traversal, and applications can use Neo4j drivers over Bolt or a typed HTTPS API.

+1,203
6

Nasiko-Labs/nasiko : One control plane that deploys, routes and secures any A2A-speaking agent
Rust
A single control-plane process, written in Rust, that sits in front of every A2A-speaking agent: it terminates TLS, authenticates each request and proxies all agent-to-agent traffic, so agents stay off the public network. Every hop gets rate limits, ACLs and tracing, with a dashboard for deploys, routing, tools and traces; bring agents written in Python, Rust, Go or TypeScript.

+1,163
7

rocketride-org/rocketride-server : Compose AI pipelines as portable JSON on a VS Code canvas, run by a C++ engine
Python
An open-source pipeline builder and runtime for AI workloads: 100+ nodes cover LLM providers, vector databases, OCR and NER, pipelines are drawn as portable JSON on a visual canvas in VS Code, and a multithreaded C++ engine runs them on your own hardware. Python and TypeScript SDKs drop a finished pipeline into an app with a few lines of code.

+1,147
8

farion1231/cc-switch : Stop editing the same model settings in several tools
Rust
A desktop control panel for the model providers, MCP servers and skills used by several AI coding tools. Useful if you alternate between Claude Code, Codex and similar tools and are tired of keeping their configuration files in sync by hand.

+1,078
9

hypit-ai/hypit : Let your coding agent turn one reference video into a hundred variants
TypeScript
A skill that gives coding agents a language for making video: drop in a reference clip and the agent rebuilds it as a complete workflow covering footage, captions, B-roll and effects, timed to words instead of seconds. You can also start from templates or a plain description, and a workflow can compile captions and motion graphics without calling a generation model.

+1,004
10

experientiallabs/experiential : Put model access, routing and budgets in one place
Python
A gateway that puts hosted and local models behind a shared API, with routing, access and spending controls. Useful when several agents need different models but you want one place to manage which services they may use and how much they can spend.

+952

Ranking window: 2026-09-23 UTC, full day.

What I would look at first

Reading notes updated 2026-09-24; the ranking above preserves its original reporting window.

Two decision models again take the top spots: NandhaKishorM/laya answers typed questions about any text in one forward pass, and jaredpalmer/kev ships the open-weights version you can train yourself. kev’s playground shows how option order shifts the probabilities. New at number three: vectorize-io/hindsight builds agent memory around three operations, retain, recall and reflect, where the reflect step keeps what an agent knows current over time. It runs embedded in Python or as a server, so testing it beside your existing retrieval setup is an afternoon’s work.

If agents are multiplying faster than your ops headcount, two entries attack that directly. Nasiko-Labs/nasiko puts one Rust control plane in front of every A2A-speaking agent: it terminates TLS, authenticates each request and proxies traffic between agents, keeping your agents off the public network. For data, hydra-db/hydradb moves a graph database onto S3-compatible object storage, with query nodes and background indexers that scale on their own; if you already write OpenCypher, its Neo4j-compatible Bolt port makes the trial cheap.

Around them, three picks by situation. farion1231/cc-switch suits anyone alternating between Claude Code and Codex who is tired of syncing config files by hand. experientiallabs/experiential gives several agents one gateway for model access, routing and budgets. google/ax fits if you want the task, the repos and the network allowlist in a single YAML file.

Decision models keep the lead, with agent memory, an agent control plane and a graph database close behind

Fastest-growing this period

NandhaKishorM/laya : Answer typed questions about any text in one 33-millisecond forward pass

jaredpalmer/kev : Small open decision models answering choice, yes/no and scoring questions in one request

vectorize-io/hindsight : Agent memory with three operations: retain, recall, and reflect to keep learning

google/ax : Declare agent tasks in YAML, sandbox them, then SSH into a running one

hydra-db/hydradb : A Rust graph database that keeps the graph on S3 and scales compute separately

Nasiko-Labs/nasiko : One control plane that deploys, routes and secures any A2A-speaking agent

rocketride-org/rocketride-server : Compose AI pipelines as portable JSON on a VS Code canvas, run by a C++ engine

farion1231/cc-switch : Stop editing the same model settings in several tools

hypit-ai/hypit : Let your coding agent turn one reference video into a hundred variants

experientiallabs/experiential : Put model access, routing and budgets in one place

What I would look at first