PERSEUS COMPUTING LLC · HIGH-ASSURANCE SYSTEMS LABORATORY · AUSTIN, TEXAS · MIT LICENSE
Flagship Product: telarch.ai ↗ · contact@perseuscomputing.com · support@perseuscomputing.com
SYSTEMS RESEARCH LABORATORY · HIGH-ASSURANCE COMPUTING

Deterministic foundations for sovereign autonomous systems.

Perseus Computing LLC is a systems research laboratory building sovereign, mathematically verified cognitive architectures. Our flagship suite, Telarch Cognitive Infrastructure (TCI), delivers hard token budgeting, encrypted local memory, and cryptographic auditability for critical enterprise and defense applications. Deterministic performance. Sub-millisecond latency. Zero cloud dependencies.

Explore Interactive Demos ↓ Read Executive Brief (PDF) ↗ View on GitHub ↗
Context Assembly
1.72 μs
Microsecond p50 execution with zero garbage collection.
Token Waste Reduction
84.8%–93.6%
Automatically packs high-priority instructions into budget.
Concurrency Throughput
1.77M+ ops/s
Multi-agent swarms with zero lock collisions or data races.
Tamper Detection
46.0 μs
Cryptographic SHA-256 logs catch modifications instantly.

How It Works: Three Core Infrastructure Engines

Instead of an opaque, all-in-one framework, the Telarch Suite divides autonomous agent infrastructure into three focused, high-assurance engines that run entirely on your own hardware.

harpe — Context Window Budgeting & Deterministic Token Pruning
NATIVE ENGINE
ENGINE GUARANTEES & INVARIANTS
  • ✓ Sub-microsecond latency: p50 execution under 1.72 μs ensures budgeting incurs zero pipeline overhead
  • ✓ Explicit pruning policies: LowestPriority, LeastRelevant, Fifo, TruncateTail
  • ✓ Zero-allocation steady state: pre-allocated token pools prevent heap fragmentation
  • ✓ Deterministic ceiling reduction up to 96.8% on standard benchmarks, up to 99.0% on full corpora
RUST INTEGRATION
VERIFIED ARCHITECTURAL FLOW

Measured on Real Hardware. Not Synthetic Marketing.

We evaluate speed, long-horizon memory recall, and concurrency using standard benchmark suites and multi-threaded stress tests—proving microsecond performance under load.

LONGMEMEVAL-S CLEANED // Needle-in-Haystack Recall Convergence
EMPIRICALLY VERIFIED

Evaluated across multi-session histories averaging 123,770 raw source tokens. Harpe compiles high-salience context down to 1,179 tokens (99.0% reduction) while preserving 98.20% Recall@10.

ARCHITECTURAL ARCHETYPES // DESIGN COMPARISON MATRIX
INDEPENDENT VERIFICATION
Capability Dimension Foundational Tooling
(Redis / SQLite)
Conversational Frameworks
(Mem0 / Letta)
Managed Agent Clouds
(Zep / Graphiti)
Paged Memory R&D
(MemScheduler / LangMem)
Telarch Suite (TCI)
(Harpe + Kibisis + Argus)
Context Ceiling Manual developer slicing Heuristic LLM-managed window Cloud heuristic extraction Token budget window paging Strict Knapsack Ceiling (0 Overruns)
Token Efficiency Application dependent Variable (tool call overhead) Graph extraction dependent Paging heuristic dependent 84.8% to 93.6% Measured Reduction
Storage & Encryption Database-level encryption Plaintext database stores Cloud database encryption In-memory / plaintext disk AES-256-GCM Envelope Encryption
Erasure Integrity Standard SQL/key delete Application-level soft delete Cloud tombstone / purge Soft delete / memory drop Cryptographic Shredding Tombstones
Multi-Agent Safety Optimistic locking / manual Process-level thread locks Server-managed transactions Single-process coordination 1.77M ops/s SWMR Monotonic Leases
Audit Provenance System transaction logs Console logs / SaaS telemetry Cloud activity history Application tracing spans 1M Scale SHA-256 Merkle Ledger
Deployment Boundary Self-hosted storage engines Developer workstations / cloud Proprietary multi-tenant cloud Application-embedded library 100% Sovereign Air-Gapped Runtimes
Compliance Readiness Host configuration required General-purpose design Commercial SOC 2 / HIPAA Unmanaged library runtime NIST SP 800-53 AU-2/3 & CMMC L2 Ready

Context Budgeting & Token Economics

See how Harpe eliminates context bloat. Choose a target token ceiling and watch it automatically pack high-priority instructions while discarding background noise—slashing API bills without losing critical details.

ALLOCATION ENGINE // KNAPSACK GREEDY OPTIMIZER NATIVE ENGINE
CEILING PRESETS:
2,048 T
1,024 T (Strict Edge) 32,768 T (32K Mid) 131,072 T (128K Frontier) 262,144 T (256K Ultra)
KNAPSACK ALLOCATION BUFFER // HARPE DETERMINISTIC PACKER 1,984 / 2,048 T (96.9% Committed)
Candidate Chunks Manifest // 275,492 Raw Input Tokens GREEDY DENSITY SELECTION [O(log n)]
Chunk Identifier Subsystem Category Raw Tokens Priority Knapsack Status Assembly Latency
Token Footprint Reduction
98.4%
62.4x Compression Ratio
Est. Cost Savings / Turn
$0.37
$3,654 per 10k Turns
Assembly Compile Latency
1.48 µs
Native Binary (Zero GC)
Context Ceiling Overrun
0.00%
Hard Mathematical Boundary

Encrypted Memory & Instant Revocation

Test client-side encryption and contradiction defense. Encrypt sensitive agent memories, acquire exclusive write locks, and verify that revoked memories can never be resurrected by stale prompts.

WEBCRYPTO ENCLAVE // ZERO-KNOWLEDGE HARDENING NIST SP 800-38D COMPLIANT
AES-256-GCM READY 1,745,280 OPS/SEC
01 // AUTHENTICATED VAULT AES-GCM-256

Client-side authenticated encryption at rest. Plaintext memory is never stored unencrypted.

[VAULT READY] Awaiting encryption trigger...
02 // SWMR CONCURRENCY LEASE MONOTONIC TOKENS

Single-Writer Multi-Reader (SWMR) lease locks. Eliminates data races across autonomous swarms.

ACTIVE LEASE STATE: IDLE (NO LOCK HELD)
FENCE TOKEN: 0x0000000000000000
LEASE TTL (50ms): 0 ms
• Readers: Unblocked (Optimistic MVCC)
• Writers: Fenced (0 Race Inversions)
THROUGHPUT: 1,745,280 OPS/SEC
03 // RECON TOMBSTONE DEFENSE 100% REJECTION

Revoking memory mints an immutable SHA-256 tombstone digest to block stale resurrection.

TOMBSTONE REGISTRY STATUS:
0 Tombstones Registered. Entity active in vault.
• 100% Anti-Resurrection Defense Active
Encryption Primitive
AES-256-GCM
128-bit MAC Authentication
SWMR Lease Throughput
1.77M ops/s
0 Data Races • Zero GC
Carrier Memory Leakage
0.0%
Volatile Buffer Physical Zeroization
Anti-Resurrection Rate
100%
0.0% Sequence Inversions

Tamper-Proof Audit Trail

Inspect a live cryptographic audit trail. Append real-time agent actions, simulate a single-bit corruption attack, and watch the verification engine catch the tampering in 46 microseconds.

PROVENANCE LEDGER // MERKLE PARENT VERIFIED AU-2/3 INTACT
CHRONOLOGICAL BLOCKS DAG 5 BLOCKS LINKED
BLOCK #02 harpe.knapsack_synthesis
SIGNATURE VERIFIED
PREVIOUS BLOCK HASH (PARENT LINK): 4a8e2f1c...
CALCULATED SHA-256 HASH: 8fa2b09c...
CANONICAL PAYLOAD (RFC-8785 JSON) 184 bytes

                
• Parent Merkle link cryptographically verified in WebCrypto (0.84 µs).
• Conforms with NIST SP 800-53 AU-2 Audit Event Generation & AU-3 Content.
TIMESTAMP: 1728584400.182 FAIL-CLOSED HALT: 46.0 µs
Hash Engine Primitive
SHA-256
NIST FIPS 180-4 Standard
Cryptographic Append Latency
4.12 µs
Pre-Allocated Ring Buffer
Tamper Detection Latency
46.0 µs
Immediate Process Boundary Halt
Chain Invariant Integrity
100.0%

Built for Defense, Enterprise, and Regulated Environments

From commercial enterprise deployments to classified defense software factories, the Telarch Suite is engineered to meet strict federal cybersecurity and auditability standards out of the box.

SPRS 110/110
NIST SP 800-171
Maximum SPRS assessment score of 110/110 with DoD CMMC Level 2 alignment and documented System Security Plans.
AU-2 & AU-3
NIST SP 800-53
Continuous cryptographic audit trail ensures non-repudiable provenance for every automated action and tool invocation.
JCP / DD2345
Defense Certified
Certified Joint Certification Program entity (0092893) approved and active through August 2031 for critical technology data handling.
MOSA COMPLIANT
Modular Architecture
Strict Modular Open Systems Approach. MIT license with zero cloud lock-in or proprietary runtime telemetry.

Open Source & Available on Crates.io

Inspect the source code, run the automated test suites, or add Telarch Suite crates directly to your projects under the MIT license.

CARGO.TOML // DEPENDENCY SPECIFICATION
RUST EDITION 2021
[dependencies]
harpe = { version = "0.1.0", path = "../crates/harpe" }
kibisis = { version = "0.1.0", path = "../crates/kibisis", features = ["subtle-crypto"] }
argus-ledger = { version = "0.1.0", path = "../crates/argus-ledger" }
• Compile time: <4.2s · Target: x86_64 / aarch64 · Deterministic verification: 100%
View Organization Repositories ↗
FREQUENTLY ASKED QUESTIONS // ARCHITECTURAL CONTEXT

Frequently Asked Questions

Key questions regarding Telarch pronunciation, etymology, sovereign air-gapped deployment, and our parent organization.

ETYMOLOGY & PRONUNCIATION
How is "Telarch" pronounced, and what does it mean?
Telarch is pronounced "tell-Ark" (/ˈtɛl.ɑːrk/). The name derives from the Greek telos (purpose, completion, or end-state) and archē (first principle, sovereign leadership, or origin). It signifies deterministic infrastructure engineered from first principles toward purposeful autonomous task completion.
ORGANIZATION & SUITE
What is the relationship between Perseus Computing and Telarch?
Perseus Computing LLC is the corporate parent and commercial research laboratory based in Austin, Texas. Telarch Suite (TCI) is the open-source cognitive infrastructure product suite (comprising Harpe, Kibisis, and Argus) published under the MIT License.
SOVEREIGN & AIR-GAPPED
Can Telarch run in disconnected or classified environments?
Yes. Telarch is engineered for 100% sovereign, local-first execution. It operates with zero external cloud dependencies, zero telemetry reachback, and zero vendor lock-in. Storage and cryptographic proofs reside entirely on your local infrastructure.
SUPPORT & INQUIRIES
How do we reach technical support or discuss acquisitions?
Direct inquiries are routed through dual monitored desks: for partnerships, defense contracting, and acquisitions, contact contact@perseuscomputing.com. For crate integration and technical architecture, contact support@perseuscomputing.com.

Direct Technical Channels

Direct communication channels for engineering teams, researchers, and technical partnerships.

ENGINEERING & INTEGRATION
Technical Support & Architecture
Direct technical access for systems engineers integrating Telarch Suite crates.
support@perseuscomputing.com →
DEFENSE & GOVCON
Teaming & Subcontracting
CAGE: 22JC5 · UEI: PJS2LW7HAK35 · Phase I/II/III subcontracting & OTA consortia.
contact@perseuscomputing.com →
SECURITY RESEARCH
Vulnerability Disclosures
PGP-encrypted disclosure channel for security researchers and auditors.
security@perseuscomputing.com →