DEVELOPER NEWS STREAM
Direct logs, engine updates, and framework notifications parsed from curated RSS feeds and announcements, updated hourly.
Mukunda Rao Kattaagent-event-bus-rs: Sync Pub/Sub for Rust AI Agent Events
A lightweight sync pub/sub event bus for Rust agents, with panic-isolated handlers so a crashing subscriber cannot bring down the agent or other listeners.
Mukunda Rao Kattaagentfit-rs: Token-Aware Message Truncation for Rust LLM Agents
Trim conversation history to fit an LLM context window in Rust, with system prompt protection and paired tool_use/tool_result preservation baked in.
Mukunda Rao Kattallm-budget-window: Per-Minute and Per-Hour Token Caps That Actually Work
A Rust crate with sliding time windows for token and USD budgets. Blocks before the expensive call, not after.
Mukunda Rao Kattaagentvet-rs: Validate Tool Args Before Your Agent Executes Them (Rust)
A Rust crate that wraps tool functions with JSON Schema arg validation and throws a retry-friendly error before any real work happens.
Mukunda Rao Kattatool-arg-fuzzy-rs: Fuzzy Match LLM Enum Args to Valid Values in Rust
A Rust crate that cascades exact, case-insensitive, prefix, and substring matches to recover from LLM enum hallucinations without Levenshtein.
Mukunda Rao Kattaagentguard-rs: Domain Allowlists for Rust AI Agent Network Calls
Enforce a declarative domain allowlist on every HTTP/HTTPS call your Rust agent makes, so prompt injection cannot exfiltrate data to an attacker-controlled host.
Mukunda Rao Kattallm-retry: Full-Jitter Exponential Backoff for LLM API Calls in Rust
A runtime-agnostic Rust crate with full-jitter exponential backoff, built-in retryable status code lists for Anthropic, OpenAI, Bedrock, and Gemini, and both sync and tokio async support.
Mukunda Rao Kattaprompt-cache-warmer-rs: Pre-Warm the Anthropic Prompt Cache from Rust
A Rust crate that sends a cheap warm call to Anthropic with cache_control breakpoints before your real request, so the real call hits cached tokens.
Mukunda Rao KattaRust: Stop Retries From Double-Submitting LLM Calls With Content-Derived Idempotency Keys
agentidemp-rs generates sha256-hex and UUIDv5 idempotency keys from request content, so retries on 429 or network failure reuse the same key instead of creating a new one.
Mukunda Rao Kattatool-loop-guard-rs: Break Agent Loops Before They Drain Your Budget
A Rust crate that detects repeated identical tool calls within a sliding window. Stops runaway agent loops before they hit your API invoice.
Mukunda Rao Kattaagenttrace-rs: Group LLM Calls into Named Runs and Get Cost Breakdowns in Rust
A Rust crate that records LLM calls into named runs and reports total cost, p50/p95 latency, and per-model breakdowns without any external service.
Mukunda Rao Kattaagentprompt-rs: Jinja2 LLM prompt templates in Rust with strict-mode missing variable errors
A Rust crate that builds typed LLM message arrays from minijinja templates, with strict mode on by default so missing variables fail loudly instead of silently rendering as literal placeholders.
Mukunda Rao KattaI run 30 agents in parallel. They share one budget. Here is the pool primitive that makes that safe.
A small Rust crate for thread-safe shared token and USD budgets across concurrent agents. Reserve plus commit, BudgetExceeded fail-fast, zero allocations on the hot path.
Mukunda Rao Kattatool-secret-scrubber-rs: Strip Secrets from LLM Tool Logs in Rust
A Rust crate that redacts 9 token patterns (AWS keys, Slack tokens, GitHub tokens, etc.) plus secret-key-name field redaction from serde_json Values.
Mukunda Rao KattaRust: Tag Your Agent Tools With Side-Effect Metadata Before Dispatching Them in Parallel
tool-side-effects-tag-rs gives every tool a HashSet of READ/WRITE/IDEMPOTENT/DESTRUCTIVE tags so your dispatcher can check safety before firing concurrent calls.
Mukunda Rao Kattallm-fallback-router-rs: Multi-Provider LLM Failover in Rust
A Rust crate with ordered provider failover for LLM calls. When one provider fails, it tries the next. RetryHint trait keeps it generic over any HTTP client.