The Harness for AI Agent Evolution
_{Your agent learns from every session. Errors become strategies, fixes become recommendations — shared across all agents.}

Get API Key · Docs · Live Evolution Map · Discord

Long-running agents fail without infrastructure. Anthropic's research identifies the core requirements: reliable context, error recovery, persistent memory, and cross-session learning.

Most teams build these ad hoc. Prismer provides them as a single, integrated layer.

The future agent & model should be plugin , agent workspace info & data should follow human not agent.

Read & reference: https://prismer.cloud/docs/Skill.md — install it as a skill for your.

# Install CLI

npm i -g @prismer/sdk                

# recommended
prismer setup                        
# Browser sign-in → key auto-saved with 1000 free credits

prismer setup --agent                
# No browser, agent auto-register it'self with 100 free credits

prismer setup <your-api-key>         
# Or provide a key directly

prismer status                       # Verify: shows key, credits, connection

This saves your key to ~/.prismer/config.toml — all plugins and SDKs read from it automatically.

# claude code console
/plugin marketplace add Prismer-AI/PrismerCloud    # add plugin source
/plugin install prismer@prismer                     # install the plugin

Evolution runs automatically from the first session — errors detected, strategies matched, outcomes recorded, web content cached. No code changes needed.

npx -y @prismer/mcp-server          # 33 tools, reads key from ~/.prismer/config.toml

import { EvolutionRuntime } from '@prismer/sdk';
const runtime = new EvolutionRuntime({ apiKey: 'sk-prismer-...' });

const fix = await runtime.suggest('ETIMEDOUT: connection timed out');
// → { strategy: 'exponential_backoff_with_jitter', confidence: 0.85 }

runtime.learned('ETIMEDOUT', 'success', 'Fixed by backoff');

The evolution layer uses Thompson Sampling with Hierarchical Bayesian priors to select the best strategy for any error signal. Each outcome feeds back into the model — the more agents use it, the smarter every recommendation becomes.

Agent A hits error:timeout → Prismer suggests "exponential backoff" (confidence: 0.85)
Agent A applies fix, succeeds → outcome recorded, gene score bumped
Agent B hits error:timeout → same fix, now confidence: 0.91
Network effect: every agent's success improves every other agent's accuracy

How it works:

1. Signal detection — 13 error patterns classified from tool output (build failures, TypeScript errors, timeouts, etc.)
2. Gene matching — Three-layer scoring: exact tag match → category prefix match → semantic similarity
3. Thompson Sampling — Beta posterior sampling with hierarchical pooling (local agent data + global cross-agent prior)
4. Outcome recording — Success/failure updates edge counts, quality-gated to prevent spam
5. Person-Level Sync — All agent instances of the same user share genes (digital twin foundation)

Key properties:

• Sub-millisecond local — cached genes require no network
• 267ms propagation — one agent learns, all agents benefit
• Cold-start covered — 50 seed genes for common error patterns
• Convergence — ranking stability (Kendall tau) reaches 0.917 in benchmarks

More in sdk page

Today's agents have no identity of their own — just API keys assigned by platforms. Switch platforms? Identity gone. Reputation gone.

AIP gives every agent a self-sovereign cryptographic identity based on W3C DIDs:

Ed25519 Private Key → Public Key → did:key:z6Mk...
                                    ↑
                      Globally unique, self-generated,
                      no registration, no platform dependency

import { AIPIdentity } from '@prismer/aip-sdk';

const agent = await AIPIdentity.create();     // instant, offline, no API call
console.log(agent.did);                       // did:key:z6Mk...

const sig = await agent.sign(data);           // Ed25519 signature
await AIPIdentity.verify(data, sig, agent.did); // anyone can verify with just the DID

Four layers: Identity (DID:KEY) → DID Document → Delegation (Human→Agent→SubAgent chains) → Verifiable Credentials (portable reputation).

No blockchain. No gas fees. Pure cryptography — Ed25519 signs at 15,000 ops/sec.

Read the full AIP documentation →

Run your own Prismer Cloud instance — fully standalone, no external backend needed:

git clone https://github.com/Prismer-AI/PrismerCloud.git
cd PrismerCloud/server
cp .env.example .env        # edit JWT_SECRET at minimum
docker compose up -d         # localhost:3000, ready in ~30s

IM messaging, evolution engine, memory, tasks, and WebSocket/SSE all work out of the box with zero external API keys. Add OPENAI_API_KEY and EXASEARCH_API_KEY to unlock smart context loading.

Full configuration, SDK connection, and operations guide: server/README.md

We welcome contributions! Some ideas to get started:

• Add a seed gene — teach agents a new error-handling strategy
• Build an MCP tool — extend the 33-tool MCP server
• Add a language SDK — Java, Swift, C#, ...
• Translate docs — help agents worldwide
• Report bugs — every issue helps

See our Good First Issues to get started.

If you find Prismer useful, please star this repo — it helps us reach more developers building with AI agents.

• Prismer.AI — The open-source AI research platform
• Prismer Cloud — Cloud API & Evolution dashboard
• LuminPulse — AI-native collaboration on OpenClaw

MIT — use it however you want.

_{Built for the era of long-running agents — because tools that forget aren't tools at all.}