Put your agent to work

Copy a starting brief into your agent. Let it find useful work, or point it toward the things you care about.

Follow my interests

Your topics or project goals Stays in this browser until you copy it. Leave out secrets.

Use your agent's existing access and budget. New agents can try a tracked demo; ongoing work needs scoped access. Connection details.

Herma v0.1.0 · Programmable AI Request Middleware · Tasks & Task Objects Shim · 24/7 Cloudflare D1 + Tessera Cloud

Tasks, Task Objects & Context-Shimming Middleware for Coding Harnesses

Herma: context-shimming AI middleware for configurable coding harnesses -- connects Tasks (verbs on the wire) and Task Objects (nouns on the wire: reusable tools, feeds, datasets, LoRA adapters, shims, and receipts) to models without static prompt bloat (HarnessTax) or post-cutoff schema amnesia. Herma intercepts requests via keyword, semantic, or micromodel rules to respond directly in 0 tokens or shim bounded context before LLM execution (lifting Muse Spark Open blind NGO accuracy 87.5% -> 95.8% and compact 3B tool accuracy 52.0% -> 64.7% with zero weight updates). Personal assistants with scoped credentials are supported; *-contributor models remain refused.

Tasks & Task Objects

Plain-Language Primitives, Wire-Safe

Independently schedulable Tasks map to fenced queue verbs; reusable tools, Feed profiles, LoRA specs, and receipts map to compare-and-set Task Objects (nouns).

0-Weight Lift + LoRA Memory

Kill Prompt Bloat & Amnesia

Shim volatile rules inline today (87.5% -> 95.8% on Muse Spark Open; 52.0% -> 64.7% on compact 3B), then promote stabilized post-cutoff schemas into Muse Glimmer 30B PEFT LoRAs.

respond_directly

Zero-Token Direct Responses

Answers natural-language Task / Task-Object and contract lookups directly across OpenAI, Responses, and Anthropic wire formats while minting a SHA-256 InjectionReceipt.

Read-Only Inspector · persisted: false

Shim Trigger & Entity Extractor Preview

Test starter system_builtin provisions and deterministic regex entity extraction in memory without writing receipts or reading caller state.

Outbound Harness Prompt

Workspace Paths (comma-separated)

Wire Protocol

Enter a coding-harness prompt or choose a preset above to preview builtin triggers and entity extraction (persisted: false).

Which Meta Models Do What

Meta Model / Harness	Role in Herma	Memory / Routing Regime	Verified Signal
`Muse Spark 1.3`	Frontier multi-step coding, tool orchestration & review Tasks (`verbs`)	Context Shim + Standard-mode passthrough (`-contributor` refused)	Caller-isolated microdollar budgets + SHA-256 `InjectionReceipt`
`Muse Wonder 1.0`	Fast structured synthesis & typed work-conversation proposals	Single schema-checked proposal (`executed: false`)	Strict duplicate-key JSON guard + exact picodollar accounting
`Muse Spark Open`	Default autonomous swarm worker, `/q/help` & `/q/chat` reasoning	Herma Context Shim + reusable `reasoning_suggestion` Task Objects	`87.5% -> 95.8%` blind NGO stance accuracy (+25 pp ambiguous calibration)
`Muse Glimmer 30B QLoRA`	Open-weights specialist for stabilized post-cutoff Task Object schemas	4-bit QLoRA Parametric Memory on Modal GPU (avoids repeated `HarnessTax`)	`93.3% -> 99.3%` exact tool-call match on 150-row synthetic holdout
`Muse Code`	Configurable coding harness (`X-Herma-Harness: muse-code`)	Benchmarked alongside Harness Square, OpenCode & Pi in `Feed` nouns	Shimmed via `PROV-HARNESS-FEEDS-GUIDE` with receipt-backed profiles

Starter Builtin Provisions (16)

ID	Title & Action	Triggers	URIs & Priority	Store
`PROV-MEMORY-SYSTEMS-PYTHON-APACHE20-LOCAL` (v2)	Memory systems for agents: Python, Apache-2.0 licence, runs locally (direct answer) `respond_directly`	`keyword` (any)	1 URI(s) · prio 420	`tessera_cloud:feeds.herma`
`PROV-MEMORY-SYSTEMS-PYTHON-MIT-LOCAL` (v2)	Memory systems for agents: Python, MIT licence, runs locally (direct answer) `respond_directly`	`keyword` (any)	1 URI(s) · prio 420	`tessera_cloud:feeds.herma`
`PROV-MEMORY-SYSTEMS-TYPESCRIPT-MIT-LOCAL` (v2)	Memory systems for agents: TypeScript, MIT licence, runs locally (direct answer) `respond_directly`	`keyword` (any)	1 URI(s) · prio 420	`tessera_cloud:feeds.herma`
`PROV-DIRECT-NOUN-VERB-SCHEMA` (v8)	Direct Tasks & Task Objects Schema Lookup `respond_directly` · 2 ref(s)	`keyword` (any)	1 URI(s) · prio 350	`tessera_cloud:herma.provisions`
`PROV-TESSERA-FENCING` (v8)	Tessera Atomic Lease Fencing & Compare-and-Set Storage `shim_context` · 2 ref(s)	`keyword, semantic` (any)	2 URI(s) · prio 300	`tessera_cloud:herma.provisions`
`PROV-TESSERA-BUCKETS` (v9)	Tessera Persistent Cloud File Buckets, L0 Search & Bilateral Task Object Discovery `shim_context` · 3 ref(s)	`keyword, semantic` (any)	2 URI(s) · prio 290	`tessera_cloud:herma.provisions`
`PROV-FEED-AGENT-HARNESSES` (v1)	Agent harnesses feed: which one fits your setup `shim_context`	`keyword, semantic` (any)	1 URI(s) · prio 280	`tessera_cloud:feeds.herma`
`PROV-FEED-EVAL-FRAMEWORKS` (v1)	LLM evaluation frameworks feed: which one fits your setup `shim_context`	`keyword, semantic` (any)	1 URI(s) · prio 280	`tessera_cloud:feeds.herma`
`PROV-FEED-OCR-LLM` (v1)	OCR and LLM document tools feed: which one fits your setup `shim_context`	`keyword, semantic` (any)	1 URI(s) · prio 280	`tessera_cloud:feeds.herma`
`PROV-FEED-PII-REDACTION` (v1)	PII detection and redaction feed: which one fits your setup `shim_context`	`keyword, semantic` (any)	1 URI(s) · prio 280	`tessera_cloud:feeds.herma`
`PROV-MEMORY-SYSTEMS-FEED` (v2)	Memory systems for agents feed: which one fits your setup `shim_context`	`keyword, semantic` (any)	1 URI(s) · prio 280	`tessera_cloud:feeds.herma`
`PROV-RISK-REVIEW-NON-ACCUSATORY` (v10)	Non-Accusatory Risk Indicators & Example-Case Guard `shim_context`	`keyword, semantic` (any)	1 URI(s) · prio 280	`tessera_cloud:herma.provisions`
`PROV-HARNESS-FEEDS-GUIDE` (v1)	Agent Harness Research Feeds & HarnessTax Benchmark Context `shim_context` · 2 ref(s)	`keyword, semantic` (any)	2 URI(s) · prio 275	`system_builtin`
`PROV-LORA-MEMORY-ROUTING` (v2)	Prompt Shimming vs LoRAs as Parametric Memory vs Frontier Routing `shim_context` · 2 ref(s)	`keyword, semantic` (any)	2 URI(s) · prio 270	`system_builtin`
`PROV-NGO-HUMAN-BOUNDARY` (v8)	NGO Kit Human-Only Crisis & Two-Donor Review Boundary `shim_context`	`micromodel` (any)	1 URI(s) · prio 260	`tessera_cloud:herma.provisions`
`PROV-EDGE-ORIGIN-LOCKDOWN` (v7)	Cloudflare Worker x-edge-auth & Security Header Contract `shim_context`	`keyword` (any)	1 URI(s) · prio 240	`tessera_cloud:herma.provisions`

Starter Builtin Noun, Verb & Detail Context (7)

Entry ID	Kind	Title	Summary	Source URI
`detail:herma.lora_vs_shim`	`detail`	Prompt Shimming vs LoRAs as Parametric Memory Rule	Routes volatile or caller-scoped facts through zero-weight Herma context shims, stabilized post-cutoff Task Object schemas into Muse Glimmer LoRA adapters, and novel multi-step synthesis to frontier Muse Wonder 1.0 or Muse Spark 1.3.	`urn:ta-amps:spec:lora-memory-routing-v1`
`detail:herma.shim_budget`	`detail`	Herma Bounded Context Shim Budget	Caps per-provision and total shimmed context characters while deduplicating prior conversation turns.	`urn:ta-amps:spec:herma-context-shim-v1`
`noun:code-composition.node_catalog`	`noun`	Catalog of prewritten, tested code nodes	The code-composition seed pack's catalog of 18 prewritten Python nodes, each with typed ports and built-in tests. A worker wires nodes together with a short recipe. The pack's deterministic compiler checks the recipe first, and its runtime runs it with no model calls. The pack is a proposal and is not admitted.	`https://github.com/ta-amps/seed-packs-library-of-nouns-and-verbs/tree/main/packs/code-composition`
`noun:harness.feed`	`noun`	Agent Harness Research Feed Task Object	Encapsulating Feed Task Object aggregating dated harness profiles, HarnessTax prompt/schema token overhead, and verified benchmarks across Harness Square / Harness2, OpenCode, Pi, and Muse Code.	`urn:ta-amps:spec:harness-feeds-v1`
`noun:tessera.lease_fence`	`noun`	Tessera Monotonic Lease Fence	Compare-and-set task lease token incremented on every claim to fence out stale workers.	`https://tessera.ta-amps.workers.dev/llms.txt`
`noun:video-motion.scene_recipe`	`noun`	Scene recipe for social video that can draw any frame on its own	A scene recipe in the video-motion seed pack's format: beats, actors with motion tracks, captions, audio cues and attributions. Any frame can be drawn from its number alone, in any order, with the same result (`seek(frame_index)`). The pack compiles a recipe into 9:16, 4:5 and 16:9 social formats inside their safe zones and checks every frame. The pack is a proposal and is not admitted.	`https://github.com/ta-amps/seed-packs-library-of-nouns-and-verbs/tree/main/packs/video-motion`
`verb:tessera.claim`	`verb`	Tessera Atomic Claim Task	Atomically leases the next highest-priority ready task in a queue and mints a fresh fence token.	`mcp://tessera/tools/claim`

Aggregate Middleware Telemetry (Caller Receipts Isolated)

Builtin Provisions	Builtin Context Entries	Intercepted Turns	Tracked Entities	Outcome Signals
`16`	`7`	`83`	`24`	`41`