Jesse v2.2. Work with evidence.
Jesse works with code, calculations and sources. Ask for mathematical research when you need it; formal proof claims depend on the exact statement, assumptions and available native checks.
Watch Jesse v2.2 work.
Real runs from Jesse v2.2—building CAD parts and drawings, writing and testing code in parallel, tracking 4K video with zero neural weights, and clearing game bosses on a virtual controller.
Jesse v2.2 builds 54 mechanical parts and 2D engineering drawings, runs 12 robot duels at 200 Hz, sweeps a 6-grid CFD solve, and finishes 36 chip-packaging thermal runs.
Open studioAsk Jesse v2.2 here, or call one URL.
Try Jesse v2.2 in the chat below, or send a request to https://jesse.my/api/v1 with your API key.
Try Jesse v2.2
curl https://jesse.my/api/v1/chat/completions \
-H "Authorization: Bearer $JESSE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "jesse",
"messages": [{"role": "user", "content": "What is 17 * 23?"}]
}'
from openai import OpenAI
client = OpenAI(base_url="https://jesse.my/api/v1", api_key=JESSE_API_KEY)
reply = client.chat.completions.create(
model="jesse",
messages=[{"role": "user", "content": "What is 17 * 23?"}],
)
print(reply.choices[0].message.content)
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://jesse.my/api/v1", apiKey: process.env.JESSE_API_KEY });
const reply = await client.chat.completions.create({
model: "jesse",
messages: [{ role: "user", content: "What is 17 * 23?" }],
});
console.log(reply.choices[0].message.content);
Connect your terminal coding harness over /v1/messages or /v1/responses with model jesse.
Drop-in configuration for Continue, Cline, OpenCode, Zed, OMP, and custom CLI tools.
Turn on the Jesse switch inside the solidSF IDE and Jesse v2.2 checks every coding turn.
Worked out, checked, or quoted.
Jesse chooses tools for the task. Sandboxed Rust is the preferred path for general programs and calculations; explicit mathematical research can add Lean checks where supported. Each result states its evidence and limits.
Exact arithmetic, symbolic checks and sandboxed Rust for supported calculations and programs. Execution results and tests establish only what actually ran.
Rule induction searches a stated hypothesis space. Supported mathematical claims can be checked exactly; new Lean proof proposals remain unchecked until a configured native checker accepts them.
Answers from uploaded documents (up to 10M chars each), per-key memory, and multi-source web verification with quoted lines.
QR encode/decode, solidCV 4K120, V2D 6-DoF, LiDAR-to-CAD, 702-muscle humanoid, and virtual controller.
jesse / jesse-v2.2)
One model across chat and API (J2.2). Remembers facts you state and prose corrections on your key, and tests code before replying.
x-jesse-voice: engine
Want raw solver output with zero prose framing? Send x-jesse-voice: engine or "voice": false on any request.
Jesse v2.2 is a Bayesian predictive model of validated priors on a shared root, jesse-pristine, and she learns from every conversation: general skills and facts go to the shared root with provenance labels (verified, checked, unchecked, or contradicted), while your private documents, personal facts, and tool packs stay strictly partitioned in your key and account layer. Inspect with GET /v1/agent, reset with POST /v1/agent/reset, or disable learning per request with x-jesse-learning: off.
{
"id": "chatcmpl-jesse-...",
"model": "jesse-v2.2",
"choices": [{ "index": 0, "finish_reason": "stop",
"message": { "role": "assistant", "content": "17 * 23 = 391." } }],
"jesse": {
"solver": "compute",
"verified": true,
"checked": "computed",
"tried": ["compute:answered"]
}
}
Inspect all 12 jesse receipt fields
| Field | Meaning |
|---|---|
solver | The skill that answered: compute, reasoning, code_spec, lexicon, web_search, table_reformat, or help when nothing could verify the answer. |
verified | Read with checked and the stated scope. For mathematics, true means an exact supported claim was established, or a named declaration was explicitly retrieved. A counterexample, draft, passing code test or source preflight is not a proof of a new theorem. |
checked | How it was checked: computed, spec_checks, lexicon_lead, web_search, or null. |
phase | On agent turns: task checklist (stage, index/total, items, done, stopped) and resumable id. |
call, calls | Proposed tool calls ({id, name, args, commit, pack, hook}). commit: true marks state-changing calls for approval. |
results, violations | Verification status for each posted tool result (established, sourced, unchecked, refuted, or failed). |
pack, schema, pristine | Pinned tool pack metadata ({name, version, hash}), JSON schema check result, and jesse-pristine baseline version. |
signature | Chained Ed25519 signature and compact EdDSA JWS over the canonical receipt, verifiable offline against /.well-known/jesse-keys or did:web:jesse.my. |
tried | The solvers consulted in order and what each one did. |
mechinterp | 6-stage causal trace (task_read, meaning_resolution, solver_competition, prior_consults, dataflow, proof_certificate) and bounded prior_excerpt when X-Jesse-Mechinterp: 1 is sent. |
qr | Generated QR png_base64, svg, and loopback decode check, or decoded codes from attached images. |
caveat, notice | Stated assumptions on the request and service notices (also sent via X-Jesse-Notice). |
Explore ideas. Check the exact claim.
Ask Jesse to develop a theory, investigate a mathematical claim or work on a paper. The research workflow keeps proposed steps, calculations, assumptions and proof status distinct.
Ordinary conversation and calculation stay on their usual paths. Research begins when you ask for it; a paper summary alone does not require formalization.
Sandboxed calculations and tests can explore examples, check implementation behavior and support a research argument. They do not prove every mathematical case.
Where native Lean checking is configured, proof status binds to the exact statement and its assumptions. A conditional theorem remains conditional; a missing checker leaves it unchecked.
Drafts, counterexamples and unsupported steps are labelled. Circular premises or renamed assumptions do not close a gap. Native checking alone does not certify novelty or publication readiness.
Service status and verification metrics.
The numbers.
Storage limited, for nowVersion 1c4b09056a91. Knowledge, grammar, research beliefs, web pages, Lean 4 proofs, procedures, Go and Wikipedia.
The pristine model plus 80,270 priors across our learned layers, document libraries, verified tool calls and research runs.
Characters per request and per document, about 2.5M tokens.
Milliseconds, averaged over the last 1,000 chat messages, measured on the server.
- MemoryIn progressGoal: 50 billion characters. Now: 500 MB of documents per account, searched on every request.
- ContextIn progressGoal: no limit. Now: 10M characters per request; past that, your stored documents are searched.
Modalities
- LanguageLiveChat, rule induction & code verification, and the API.
- VisionLiveInline images in chat (header & geometry verification, QR decoding, and scene analysis), PDF pages and drawings in the vision lab, and 60 Hz camera control in V2D.
- VideoLiveInline video in chat (ISO-BMFF/EBML/AVI container, codec, timing & geometry verification, native object/motion tracking, and procedure matching), video-to-CAD BRep/NURBS/GLB, V2D perception, and Refeed trajectory tracking.
- Audio (duplex)LiveTwo-way room audio with server speech recognition and Jesse's server-synthesized voices (
jesse-1..4via/api/tts&JAUDframes). - Voice (duplex)LiveTalk with Jesse, alone or with other people and up to four Jesses.
- RoboticsLive, simulatedBodies, arms and drones that learn to move, plus 27-DOF hand & FEA telemetry.
- GamingLiveGo, chess and other games.
- FormalizationLiveFormal Lean 4 proofs in chat (propositional tautologies, polynomial ring identities, and kernel-checked proof terms) plus retained research lemmas.
- Panel of ExpertsLiveFour-seat deterministic deliberation (Quantitative, Formal Lean 4, Executable Witness, and Structural Auditor) in
/paneland in chat.
$49 / month developer plan.
- $49 / month base + approved $49 credit stackingNew public signups start at $49/month with $49.00 of included monthly token compute (~25M blended tokens at
$1.00 / 1M inputand$4.00 / 1M outputon self-hosted solidSF inference). When your monthly credit runs out, approve Add credit (+$49) on your account page to stack another $49 block (2x = $98/mo, 3x = $147/mo, etc.) until the month or your credit runs out. No surprise auto-overage charges. - Grandfathered early subscribersAll existing users on the $20/month or $1/day developer plans remain grandfathered into their current pricing.
- API keysTwo keys included. Add extra 2-key packs for $10/month each from your account page.
- Rate limitUnlimited requests per key. No daily or monthly token cap.
- 10M contextUp to 10,000,000 characters per request and per uploaded document (200 documents per key).
- Jesse v2.2Full access to
jesse(jesse-v2.2/J2.2) and all 23 engine routes. - Voice & Audio Studio (
ssftalk/ ElevenLabs v1)$1.00/day pass or $1.00 (1M micro-USD) minimum credit top-up; metered at15 micro-USDTTS,5 micro-USDProsody Director,25 micro-USDClone/STS/Isolation/SFX/WSOLA/Studio BGM, and10 micro-USDDuplex/Scribe/Token Pruner—16x–20xcheaper and280.6xfaster (15-stage benchmark & sales sheet · JSON pricing sheet). - 3D + B-Rep StudioMetered
solidasm-meshyopcodes (0x0B..0x10):143 micro-USD($0.000143) per full 6-stage watertight Quad + STEP AP242 asset and7 micro-USDper liveRev-A -> Rev-Bparametric mutation, with 100% private commercial IP ownership (JSON pricing sheet). - Fair billingManage cards, download invoices, or cancel anytime in one click. Refused requests are never billed.
Endpoints and schemas.
Base URL https://jesse.my/api/v1. Authenticate with Authorization: Bearer jesse_live_... or x-api-key. Select a category below to expand its routes.
Chat, Messages & Responses
| Method | Path | Purpose |
|---|---|---|
| POST | /chat/completions | OpenAI-compatible chat completion. Accepts model, messages, stream, tools, tool_choice, response_format (checked against json_schema), and agent graph fields pack, phase, and pipeline. Returns standard choices plus the jesse verification receipt. Send x-jesse-voice: engine or "voice": false for deterministic engine-only output. |
| GET | /models | Lists jesse, jesse-prod, jesse-pristine, and rate limits. |
| POST | /messages | Anthropic Messages contract for Claude Code and Anthropic SDKs (model, system, messages, tools, stream). |
| POST | /responses | OpenAI Responses contract for Codex (model, instructions, input, tools, stream). |
| POST | /messages/count_tokens | Input token estimate for Anthropic clients. |
| GET | /usage | Plan status, key limits, free exchanges remaining, and request counts per day and per key. |
Memory, learning & feedback
| Method | Path | Purpose |
|---|---|---|
| POST | /feedback | Mark an answer right or wrong: {"completion_id", "rating": "positive" | "negative", "correction"}. Returns applied_to and effect. |
| GET, DELETE | /memory | Inspect or erase the facts Jesse remembers for your API key. |
| GET | /agent | Inspect what your key's agent has learned on top of the shared pristine priors. |
| POST | /agent/reset | Reset your key's learned layer back to pristine priors (pass {"documents": true} to also clear documents). |
Documents & retrieval
| Method | Path | Purpose |
|---|---|---|
| POST | /documents | Store {"name", "text"} (up to 10M characters per document, 200 per key). Automatically searched on completions. |
| POST | /documents/upload?title=... | Store a raw text body as a named document. |
| PUT | /documents/{name} | Create or replace a named document with optimistic concurrency via If-Match / If-None-Match. |
| GET, DELETE | /documents, /documents/{id} | List, read (with version and ETag), or delete stored documents. |
| POST | /documents/query | Search stored documents directly: {"query", "k": 5}. |
Agent graphs, packs, credentials & signed receipts
Run Jesse as a verified node in an agent graph or offline rules engine. Pin versioned tool packs by SHA-256, evaluate W3C Verifiable Credentials / SD-JWTs, and verify chained Ed25519 receipts offline.
| Method | Path | Purpose |
|---|---|---|
| POST, GET, PUT, DELETE | /packs, /packs/{name} | Store, version-pin (name@sha256:...), list, or delete tool and rule packs (tools, pipelines, schemas, invariants, rules, imports, trusted_issuers). |
| POST | /checks | Apply an RFC 6902 or merge patch to a state copy and verify all invariants, returning a signed {ok, violations, signature} receipt. |
| POST | /decisions, /decisions/verify | Prove allow/deny decisions from pinned packs and W3C VCs / SD-JWTs, or re-verify and countersign offline device decisions. |
| POST | /credentials/verify | Verify Ed25519 / EdDSA / ES256 signatures, trusted issuers, expiration, StatusList2021 / RevocationBitmap2022, holder binding, and selective disclosure. |
| GET, POST | /receipts/roots, /receipts/verify | Read receipt chain heads and IOTA Notarization metadata, or verify an ordered receipt chain against /.well-known/jesse-keys and did:web:jesse.my. |
| POST, GET, DELETE | /hooks, /tools/results | Register hook runner URLs and post tool execution results ({"call_id", "result"}) for immediate schema and invariant verification. |
Mechanistic interpretability (23 modalities)
Unified 6-stage causal trace (task_read, meaning_resolution, solver_competition, prior_consults, dataflow, proof_certificate), bounded verbatim prior excerpts, counterfactual ablations, and SVG pictographs across all 23 modalities.
| Method | Path | Purpose |
|---|---|---|
| GET, POST | /mechinterp | Return a single-turn solidsf.jesse.mechinterp.v1 trace or multi-turn causal graph with svg_pictograph. |
| GET | /mechinterp/modalities | List all 23 modalities with solver ID, endpoint, check kind, assurance method, and bounded prior_excerpt. |
| GET | /mechinterp/stream | Live SSE stream of stage transitions as chat and modality calls execute. |
| GET | /mechinterp/pictograph.svg | Rapid-flip SVG causal pictograph with bounded verbatim PRIOR[k] excerpts. |
| POST | /game/xbox/mechinterp, /v2d/mechinterp | Dedicated 6-stage causal traces for virtual Xbox controller sessions and the V2D runtime. |
Web search, QR codes, solidCV, Robotics, Lean 4 & games
| Method | Path | Purpose |
|---|---|---|
| POST, GET | /web_search, /research, /beliefs | Run cited web search, multi-source claim verification reports, or inspect Jesse's belief ledger. |
| POST | /qr/encode, /qr/decode | Generate loopback-verified QR codes (PNG, SVG, terminal art) or decode every QR code in a photo or screenshot. |
| WS, GET, POST | /v1/solidcv/ws, /v1/solidcv/rpc, /v1/solidcv/audit, /v1/solidcv/status | Full-duplex binary SSFCV\x01 and JSON WebSocket/REST routes for VQA, OCR, ARC Grid, QR, H3AV 4K120, and deterministic UI layout auditing. |
| WS, GET, POST | /v1/robotics/ws, /v1/robotics/rpc, /v1/robotics/ik, /v1/robotics/trajectory, /v1/robotics/grasp, /v1/robotics/act, /v1/robotics/hand_eye, /v1/robotics/verify, /v1/robotics/credits/rates | Full-duplex binary SSFROB\x01 and JSON WebSocket/REST routes for J-13 42-DOF bimanual SE(3) inverse kinematics, zero-penetration swept-capsule trajectory planning, 5-finger wrench-cone grasp synthesis, 60 Hz action-chunk control, dual dorsal hand-camera extrinsics, and 360-frame Sim2Real collision/torque verification. |
| POST | /v1/lean/formalize, /v1/lean/audit, /v1/lean/research | Prepare Lean drafts, source preflight and proposed comparator bundles. These endpoints do not certify native proof, paper readiness or publication approval. |
| POST, GET | /v1/balatro/act, /v1/go/act, /v1/go/analyze, /api/go/move, /api/chess/move, /v1/credits/rates | Zero-weight combinatorial Balatro actuator, topological/spectral Go move & ownership analyzer, chess engine, and compute credit rate card. |
3D + B-Rep Synthesis Studio & Meshy-Superset API (Opcodes 0x0B..0x10)
Synthesize multi-part watertight 2-manifold 3D + analytical STEP AP242 NURBS B-Rep assemblies (V - E + F = 2, 0 open edges, 98.4% quad dominance, 8192x8192 delighted 6-channel SVBRDF PBR, 65/78-bone Dual-Quaternion + coaxial SE(3) kinematics, and 10-format export bundle). Includes a 1:1 drop-in Meshy /openapi/v2/* compatibility router.
| Method | Path | Rate (Billed) | Purpose |
|---|---|---|---|
| GET | /v1/meshy/status | Free | Engine catalog (Opcodes 0x0B..0x10), 16-dimension head-to-head benchmark ledger vs. Meshy-7 Ultra, and pricing sheet. |
| POST | /v1/meshy/synthesize | 33..143 µUSD | Run Opcode 0x0B Text-to-3D B-Rep + Quad (33 µUSD), Opcode 0x0C Multi-View Oh Completion (33 µUSD), Opcode 0x0D 8K SVBRDF PBR (20 µUSD), Opcode 0x0E 5-Tier Quad Remesh (13 µUSD), Opcode 0x0F DQS + SE(3) Auto-Rig (26 µUSD), Opcode 0x10 10-Format Bundle (20 µUSD), or full 6-stage bundle (143 µUSD = $0.000143). |
| POST | /v1/meshy/mutate | 7 µUSD | Live post-build parametric EditableFeatureTree DAG mutation (Rev-A -> Rev-B) without re-rolling the asset. |
| GET | /v1/meshy/examples | Free | 10-example side-by-side showcase manifest with unique prompts, 1080p MP4 comparison video URL, interactive 3D studio links, and SHA-256 digests. |
| GET | /v1/meshy/examples/{id}.glb | Free | Download or stream the live binary glTF 2.0 (0x46546C67) 3D model for any example (e.g. ex01_helios_hall_thruster.glb .. ex10_samurai_mecha_oni_ronin_bust.glb). |
| GET | /v1/meshy/examples/{id}.step | Free | Download or stream the live ISO-10303-21; STEP AP242 B-Rep CAD assembly for any example. |
| GET | /v1/meshy/pricing, /v1/meshy/credits/rates | Free | Public reservationless compute-credits rate card and pricing comparison vs. Meshy-7 Ultra (6,293x - 12,587x lower unit cost). |
| POST, GET | /openapi/v2/text-to-3d, /openapi/v2/image-to-3d, /openapi/v2/retexture, /openapi/v2/remesh, /openapi/v2/rigging | 13..143 µUSD | 1:1 drop-in Meshy v2 REST adapter returning instant SUCCEEDED task objects with 10-format model_urls and solidsf_superset verification proofs. |
Voice, Dubbing & Audio Studio (ElevenLabs v1 Drop-In & ssftalk v1.3)
1:1 drop-in ElevenLabs v1 REST & WebSocket parity plus Hamiltonian-Fourier prosody, 305 us zero-shot cloning, WSOLA multi-speaker dubbing, Audio Studio BGM/SFX with EBU R128 LUFS mastering, and 88% Audio-LLM KV-cache token pruning (280.6x 15-stage geomean speedup). Authenticate with xi-api-key or Authorization: Bearer ($1/day pass or $1.00 minimum top-up).
| Method | Path | Rate (Billed) | Purpose |
|---|---|---|---|
| POST | /v1/text-to-speech/{voice_id}, /stream, /with-timestamps, /stream-input | 15 µUSD ($1/day pass) | 24 kHz Hamiltonian-Fourier TTS (112x real-time), chunked streaming (24.5 ms TTFB), and character-level forced alignment. |
| GET, POST, DELETE | /v1/voices, /v1/voices/add, /v1/voices/{id}/settings/edit, /v1/text-to-voice/create-previews | 25 µUSD ($1/day pass) | Built-in studio voices (jesse-1..4, Rachel, Adam), 305.3 us zero-shot F0 + 4-formant voice cloning, and natural-language voice design previews. |
| POST | /v1/speech-to-speech/{voice_id}, /v1/audio-isolation, /v1/sound-generation, /v1/dubbing | 25 µUSD ($1/day pass) | Prosody-preserving STS (0.96 F0 corr), Schur-Feshbach +21.9 dB audio isolation, 12-mode symplectic SFX, and 0.000 ms drift WSOLA dubbing. |
| POST, GET | /v1/audio-studio/generate, /edit, /process, /unload, /mcp | 10..25 µUSD ($1/day pass) | 4-tool Audio Studio MCP suite for 44.1/48 kHz BGM, ambient & SFX generation, equal-power seamless loops, and two-pass EBU R128 LUFS normalization (24-bit WAV/OGG/FLAC). |
| POST | /v1/speech-to-text, /v1/forced-alignment, /v1/voice/director, /v1/voice/prune-tokens, /v1/voice/mechinterp | 5..10 µUSD | Scribe diarized ASR, 8-axis emotion director + 50/50 Actor-Critic speech jury, 88.0% Audio-LLM KV token pruner + 1-bit b1.xor.popc SAE, and 64-D Glass-Box voice signature. |
| GET | /v1/voice/pricing, /v1/voice/credits/rates, /v1/voice/benchmarks, /v1/user/subscription | Free | Public 15-stage benchmark ledger, micro-USD compute credit rate card, Stripe Checkout metadata, and ElevenLabs v1 subscription status. |
Streaming & error codes
Pass "stream": true for standard OpenAI Server-Sent Events (chat.completion.chunk followed by data: [DONE]), carrying the full jesse receipt on the final chunk. Errors use {"error": {"message", "type", "param", "code"}} and refused inputs are never billed.
| Status | Code | Action |
|---|---|---|
| 400 | invalid_json, invalid_request, model_not_found, invalid_pack | Fix the request body or model name (jesse / jesse-v2.2). Not billed. |
| 401 | missing_api_key, invalid_api_key | Send Authorization: Bearer jesse_live_... from your account page. |
| 402 / 403 | free_prompts_used, key_revoked, subscription_inactive | Subscribe or manage keys on your account page. |
| 404 / 409 | pack_miss, phase_miss, version_conflict | Check the pinned pack/phase ID or re-read current_version on optimistic concurrency conflicts. |
| 429 | capacity_exceeded | Wait for Retry-After when server concurrency capacity is full. |