Piggy was a hand-rolled OpenAI tool loop. It is now a Prime Agent session —
Prime Intellect's own harness, embedded as a Node library — answering from
PIG's tools and, for the first time, able to put information into the CRM
rather than only read it out.
The harness is a coding agent, so the first job was taking the coding agent
away from it. `noTools: 'all'` plus an explicit allowlist leaves the model
with PIG's ten `pig_*` tools and no bash, no filesystem, no IPython. That
holds under attack: a hostile extension, a skill and a settings file planted
in the agent's own directory, then `setActiveToolsByName` called with every
built-in, still leaves ten tools, all ours. Both lines are load-bearing —
`noTools` alone registers nothing, and the allowlist is what admits our own.
Writing is gated rather than assumed. A change is proposed, not made: the
tool returns a description, the transcript renders a diff card, and nothing
reaches the database until someone presses Apply. Contracts, commitments,
allocations and compliance always stop for a human whatever the mode. Every
write runs through `executeMutation` as the calling user, so their
capabilities and the audit trail apply exactly as they would to a human's.
Four things about the SDK are wrong in its own documentation and cost a
debugging cycle each: models.json does not resolve an env var name for
`apiKey`, it sends the literal string; there is no built-in prime-inference
provider in 0.84.1; a ResourceLoader you pass in is never reloaded for you;
and the stock system prompt is a coding-assistant prompt that must be
replaced — but replacing it also silently removes the tool list, because the
harness only renders that section when it owns the prompt. AGENTS.md records
all four.
The expensive one was thinking level. The harness defaults to `medium`, and
nemotron spent an entire 4,096-token budget reasoning and returned an empty
answer. `low` was worse; `off` omits the parameter so the endpoint's default
wins. An explicit `reasoning_effort: none` via `thinkingLevelMap` took a turn
from 6,195 output tokens to 149.
And a turn is now bounded. The harness loop is `while (true)` with no
iteration cap; a runaway on a frontier model would have eaten the credit it
is supposed to report on. Ceilings on model calls and tokens, enforced both
through the harness hook and independently from the event stream, plus a
per-user daily spend limit — and the ledger now records spend on turns that
fail, which it previously discarded.
Signing in lands on /piggy, which is a workspace: conversations down one
side, the agent in the middle, what it did and what it cost beside it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Nothing in PIG derives from that repository. The fact model, the leased
agent task queue and the `agentBrief` field are our own designs, and MIT's
attribution condition reaches copied source, not ideas — so the credit was
a courtesy that misstated where this code came from.
The one line worth keeping was never a credit: AGENTS.md's rule against
lifting component files out of somebody else's repo. It is restated
generically, and the shadcn-from-upstream guidance stays.
Buzz keeps its NOTICE entry.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
PIG is an agent-native CRM for two-sided AI-compute companies: businesses
that buy GPU capacity from providers and resell it. Their business is the
spread between two pipelines, which is precisely what a generic CRM cannot
represent.
The load-bearing decision is the `allocations` table, joining a
capacity_commitment (what we bought, at a known cost) to a demand_deal
(what we sold, at a known price). Margin, utilisation and idle capacity all
fall out of that one join. Cost is charged against the full commitment
rather than only the hours that sold, because unsold hours are already paid
for and any other treatment flatters a block that is losing money.
Domain decisions worth noting, each grounded in how this market operates:
- Demand stages put `legal` second, not last. Customers do not hand
workloads to an infrastructure provider before paper is executed.
- Supply qualification splits technical from financial diligence, recorded
attributably. Accepting capacity is a two-key decision.
- Capacity carries a time SHAPE (intervals + quantities), not a window.
Commitments ramp and step down; a rectangle reports availability that
does not exist in the month someone wants it.
- SLAs model three distinct shapes: none, a reliability tier plus credits
policy, and a negotiated agreement. Aggregators generally cannot promise
uptime on resold capacity, but negotiate heavyweight paper upstream.
Remedies include fee abatement, which is materially better than a capped
credit and is not expressible as one.
- Export control is a predicate on the allocation edge, evaluated against
the ULTIMATE parent's jurisdiction. Country of incorporation is not a
valid key, so this cannot live as a flag on an account.
- Agent-derived claims land in `facts` with a confidence band and evidence.
Only verified claims self-apply; weaker ones await review.
- The API never calls the agent. It writes to a leased queue, guarded by a
partial unique index on unfinished work.
Verified: typechecks clean, migration generates and applies to Postgres 16
(31 tables, 24 enums, 117 indexes).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
claude
karti