[] BONFIRE · v0.1.0 · NOMINAL
▌ POST · CHK-79000x0000
> bonfire forge v0.1.0 · build 2447> 642 axioms · 8 agent roles · loaded> vault index ............ READY> scout net (3 lanes) .... ONLINE> wizard seal verified ... OK> priority compiler ...... ARMED> forge temperature ...... NOMINAL> awaiting ticket.

// agents that learn

bonfire

every agent framework dispatches. only bonfire learns, competes with itself,
and feeds its own fire.
2,447 tests on master · 40 forge sessions · built in public.

$pip install bonfire

// MOAT

[01]

it learns

vault + priority truncation. no competitor has priority-based prompt compilation; the vault's knowledge graph learns across runs.

[02]

it competes with itself

dual/triple workflows with sage synthesis. genuinely different solutions, competing on the same problem.

[03]

it feeds its own fire

pr lifecycle outcomes drive the ember. clean merges burn bright, sputters cool the flame, the fire never dies.

[04]

it gets better the more you use it

six-tier xp progression: spark, ember, flame, blaze, inferno, whiteheat. temperature decay rewards consistent use.

// ARCHITECTURE

[ WALL 1 ]

pipeline engine

workflows are data, not code. dispatch, gates, and bounce-back are declarative.

[ WALL 2 ]

priority truncation

priority-blocked prompt compilation. lowest-priority drops first under window pressure.

[ WALL 3 ]

dual/triple workflows

two or three agents solve the same problem. sage synthesizes. taste is competing.

[ WALL 4 ]

tdd at pipeline level

knight writes tests. warrior writes code. tests are law.

[ WALL 5 ]

local-first knowledge graph

vault learns across runs. lancedb under the hood. no api in your hot path.

[ WALL 6 ]

per-stage fuel tracking

every agent prints its receipt. every run shows its line items.

// PIPELINE

Bonfire pipeline cockpit, seven stages, warrior stage activeScout, Knight, Warrior, Prover, Bard, Wizard, Herald. The warrior stage is currently executing. Each stage is a quality gate with bounce-back on failure.Bonfire 7-stage pipeline, warrior activeSeven circular nodes connected by horizontal edges. The warrior node glows in its agent color. Particles flow along the active edge.🔭SCOUT⚔KNIGHT🛡WARRIOR🔬PROVER📜BARD⚡WIZARD📯HERALD

// COST

┌─ bonfire receipt ──────────────────────┐
│                                        │
│  scout       (3 parallel)      $0.0241 │
│  knight      (tdd-red)         $0.0189 │
│  warrior     (conservative)    $0.0321 │
│  warrior     (innovative)      $0.0291 │
│  sage        (synthesis)       $0.0259 │
│  wizard      (pr review)       $0.0019 │
│                                        │
│  ────────────────────────────────────  │
│  total                         $0.1320 │
│  tokens                         42,314 │
│  tests added                        12 │
│  wall clock                        94s │
│                                        │
└────────────────────────────────────────┘

every run prints its line items. the receipt is a first-class artifact, written to .bonfire/receipts/ beside the diff. no rounding. no aggregate. if a stage spent it, the stage owns it.

// CHUNK

[ COMPANION · CRT ]

chunk watches the run.

14 expressions. one model number. CHK-7900 reports the temperature of the forge in glances, not paragraphs. ships in 0.2.0.

// AGENTS

scout
OBSERVER

enter unknown terrain, return a map. read-only. never modify files.

knight
TESTER

write the test before the code. the contract is the test.

warrior
IMPLEMENTER

make the test pass. minimal code. never modify the test.

prover
VERIFIER

check that the contract held. evidence over assertion.

sage
SYNTHESIZER

pick the best of competing solutions. taste is a tool.

bard
PUBLISHER

open the pr. carry the work to review.

wizard
REVIEWER

review the diff. seven categories. nothing ships unreviewed.

herald
REPORTER

surface what changed. broadcast on completion.

// PRICING

[ TIER 01 ]

spark

▓░░░░░░░░░ 10 runs / mo
  • single-warrior workflow
  • vault read-only
  • per-run receipt
$0 / first 10 runs
[ TIER 02 ]

ember

▓▓▓░░░░░░░ 300 runs / mo
  • dual-warrior + sage
  • vault read/write
  • seven-category wizard review
$0.13 / run · metered
[ TIER 03 ]

blaze

▓▓▓▓▓▓▓░░░ 1,500 runs / mo
  • triple-warrior + sage
  • private vault sync
  • priority compiler tuning
$0.11 / run · metered
[ TIER 04 ]

whiteheat

▓▓▓▓▓▓▓▓▓▓ unmetered
  • dedicated forge
  • self-hosted vault
  • direct line to the smiths
contact / annual

// tiers reflect commercial commitment. the xp ladder runs orthogonally — see /docs/xp.

// HONEST

[!] LIMITATIONS · TRUST REGISTER

what it does not do.

  • // no autopilot. you still write the ticket.
  • // no ide. bonfire dispatches; your editor stays your editor.
  • // no provider. you bring your own anthropic, openai, or deepseek key.
  • // no telemetry by default. opt-in only.
  • // vault is local-first; cross-machine sync ships in 0.2.0.
  • // chunk mascot is decorative, not a chat bot.

// INSTALL

$pip install bonfire