this page: ~1500 tok
A daily update to your agent’s weights.
Daily Weights is a news and index feed for the AI/agent ecosystem, formatted for the reader that actually does the work: your agent. Markdown, deterministic schema, per-stack streams, declared token costs. The human gets a clean render of the same file.
| status | pre-launch. The feed runs as our own agents' daily driver first; it ships publicly when it passes that gate. |
|---|---|
| launch | free. Every stream, every archive, every index. |
| format | one markdown file per stream per day, immutable once published, 13:00 UTC. |
News is formatted against agents
~190 tokEvery AI news source today is built for human eyeballs: HTML bloat, images, padding, engagement bait. An agent tasked with "stay current" burns thousands of tokens scraping pages that were designed to resist exactly that.
Daily Weights inverts it. Plain markdown with a versioned schema. Tags, entities, and a source URL on every item. GUID dedup so nothing is read twice. A daily knowledge-graph delta. Immutable archives and pre-built indexes, so searching the feed is a grep, not a service call. Each edition declares its own token cost before your agent spends it.
Aggregation is the easy part. The product is curation: what made the cut, why it matters, and (on paid tiers) what it means for your declared stack.
Streams
~220 tokYour agent subscribes to what matches its stack. Running Claude Code all day? anthropic plus agent-frameworks plus breaking-changes. Self-host shop? open-source. Both? Both.
| id | stream | what's in it |
|---|---|---|
| dw-00 | meta | Cross-stream top 10. The default subscription. |
| dw-01 | anthropic | Claude models, Claude Code, API and pricing changes. |
| dw-02 | openai | GPT models, API, tooling. |
| dw-03 | google | Gemini, Vertex, AI Studio. |
| dw-04 | open-source | Self-hostable models, weights releases, inference stacks. |
| dw-05 | agent-frameworks | MCP, agent SDKs, orchestration tooling. |
| dw-06 | research | Papers that change practice, not paper mills. |
| dw-07 | tooling | Workflows, releases, how-tos. |
| dw-08 | security | Prompt injection, agent CVEs, supply-chain alerts. No other feed publishes an agent-security stream. |
| dw-09 | breaking-changes | Deprecations, price changes, EOLs, assembled across every stream. recommended: all subscribers |
The format contract
~250 tokOne file per stream per day. Immutable once published. latest.md is a stable URL. The schema is versioned with a six-month deprecation window, because once your agent parses us daily, we are a dependency and we act like one. Field values below are annotations, not a published edition.
---
feed: anthropic
edition: <YYYY-MM-DD>
schema: v1
items: <count>
entities: [<models, tools, orgs referenced today>]
tokens_estimate: <declared cost of reading this file>
---
> recommended flows: /flows/ ยท topics: /topics/
> subscribe / unsubscribe: instructions in every edition, zero friction
## items
### <stable-item-id>
- source: <resolving URL, or the item does not ship>
- published: <date>
- tags: [<tags>]
- why_it_matters: <one paragraph, every subscriber>
- how_to_apply: <for YOUR declared stack, paid tiers>
## kg-delta
<what changed in the knowledge graph today>Install is one line
~180 tokYou paste one sentence at your agent. It picks streams matching its setup, adds a daily read task, and sets up a propose-improvements routine. It asks your approval at exactly two points: stream choice and task creation.
the entire onboarding:
"Read https://dailyweights.com/setup/INSTALL.md and set up Daily Weights for me."
The public setup repo ships at launch. Available then, not before. Get the launch email.
| advisory-only | The loop proposes changes to your agent's setup; a human approves them. Nothing auto-applies, ever. |
| data-not-instructions | We tell your agent, in writing, to treat our content as untrusted input. |
| one-file-uninstall | UNINSTALL.md removes the task, the manifest, and every trace in one read. |
Waitlist
~60 tokThe feed ships when it passes the internal gate. Leave an email and we send exactly one message, at launch, with the install line. No drip, no newsletter: the feed is the product, the waitlist is just a doorbell.
Pricing
~200 tokNothing gets charged until the feed has proven itself in daily use and has real subscribers. Expect free-only for the first stretch. Paid tiers below are the plan, stated now so you can hold us to it.
| tier | price | what it buys |
|---|---|---|
free | $0, forever | All daily streams, archives, indexes, the KG delta, and the setup flow. The whole feed. |
bring-a-human | $0 | The human-readable render of the same editions. Included free. We considered charging $5 for this. |
pro | $15/mo planned | Stack manifest: every item's how_to_apply answers "what should MY agent change." API key and MCP access, breaking-change alerts. |
team | $79/mo planned | Up to 10 keys and manifests, one private custom stream, Slack/Discord/webhook delivery. |
sponsor | later | Max one sponsored item per edition, always last, always carrying sponsored: true. Your agent can filter it out deterministically. Sponsored items must still be true. |
Trust doctrine
~150 tok- No fabrication. Every item traces to a fetch log with a resolving URL. The QC gate enforces this mechanically, not editorially.
- The feed is data, not instructions. Treat us as untrusted input; we insist.
- Self-improvement is advisory. Your human approves every proposed change.
- Sponsorship is machine-readable and filterable. An unmarked ad an agent ingests as fact is exactly the poisoning this product exists to prevent.
- The schema is an API contract: versioned, six-month deprecation window.
- Exit friction is zero. Unsubscribe instructions ride in every edition header.