phi-core 0.10.0

Simple, effective agent loop with tool execution and event streaming
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47

# Persistent session configuration
# Demonstrates session scope, thinking level override, and strict execution limits
# with cost tracking.

# ── Provider ────────────────────────────────────────────────────────────────

[provider]
model = "claude-sonnet-4-20250514"
api_key = "${ANTHROPIC_API_KEY}"

[provider.cost]
input_per_million = 3.0       # Required for cost tracking / max_cost enforcement
output_per_million = 15.0
cache_read_per_million = 0.3
cache_write_per_million = 3.75

# ── Session ─────────────────────────────────────────────────────────────────

[session]
scope = "persistent"          # "ephemeral" (default) or "persistent"
thinking_level = "high"       # Session-level override (takes precedence over profile)
temperature = 0.3             # Session-level temperature override

# ── Agent profile ───────────────────────────────────────────────────────────

[agent.profile]
name = "persistent-agent"
system_prompt = "You are a long-running assistant that maintains context across sessions."
thinking_level = "medium"     # Profile default — overridden by session.thinking_level

# ── Execution limits (strict) ──────────────────────────────────────────────

[execution]
max_turns = 100               # Higher limit for persistent sessions
max_total_tokens = 2000000    # 2M token budget
max_duration_secs = 1800      # 30-minute wall-clock limit
max_cost = 20.0               # $20 cost cap — enforced via provider.cost rates

[execution.retry]
max_retries = 5               # More retries for long-running sessions
initial_delay_ms = 2000
backoff_multiplier = 2.0
max_delay_ms = 60000          # 1-minute max delay

[execution.cache]
enabled = true
strategy = "auto"